Memia #2026.27: AI vampires🧛 UWORLD U1🤖💙 compression for human🗜️🧠 j-lens🪞 orbital data embassy🛰️💾 AI allegory steganography📖 summer of ludd📵 tinyretropad📝 balcony solar 🌅🔌 stormwall☀️🛡️

Found my tribe🤖 🤖

Jul 07, 2026

Welcome to Memia #2026.27… another weekly full-spectrum scan across the frontier of AI, emerging tech and the accelerating future. As always, thanks for reading!🙏

I’m away on a bit of a break this week, exploring the West Coast of Te Wai Pounamu here in Aotearoa. This week we have had storms and massive flooding on the East Coast just over 100km away, but over here on the other side of the mountains it is unseasonably warm and dry for mid-winter (and windy!). Here’s last night’s sunset in Kawatiri Westport harbour, spectacular cloud formation coming inland from Te Moana-a-Rehua:

ℹ️PSA: Memia sends *very long emails*, best viewed online or in the Substack app.

This newsletter is now largely produced using Memia Sensorium, this week using AI agents to ingest 60+ technology news feeds, scan ~3446+ articles and curate / collate / summarise ~207 future signals below.

AI, so: expect errors, hallucinations. Please click through to read the original linked articles that interest you - images are generally those included in the RSS syndication feeds of the articles being linked to.

To keep up with the latest signals that Memia is tracking in real time, you can follow the combined Sensorium Feed at https://www.memia.com/insights/signals/ …

Experimenting with another format change again this week to make the newsletter more digestible: some sections are abbreviated with just links to the originals.

Reminder: Paid subscribers can access the full newsletter in Sensorium Beta, including the entire searchable Memia newsletter archive and the ability to dig into deeper AI analysis of this week’s stories here.

(Email me for an invite code if you haven’t already).

🗞️Weekly roundup

The most clicked link in last week’s newsletter was the excellent mind-expanding presentation by Michael Florea: Plan A: How Humanity Can Survive AGI Through Neuroengineering.

Found my tribe🤖 🤖

I had a great time hanging out with the AI Acceleration Centre team at Auckland University of Technology robotics lab last week. Thanks for hosting me Abby Dowd, Asher Finlayson, Dan Strawbridge and Alex Hiramatsu. Great to meet the rest of the team too:

No alternative text description for this image

🥄💀📈Bending Spoons, extracting rent

Bending Spoons: The zombie whisperer defying the SaaS apocalypse
Milan-based Bending Spoons, which acquires, cost-strips and “revitalises” stagnating software brands like Evernote, AOL, and Vimeo, surged 40% in its US market debut—closing at US$40.50 and achieving a US$25.7 billion market cap, more than double its last private valuation. The company’s Q1 results on the face of it tell a remarkable story: US$601 million in revenue and US$27.4 million in net income, compared to a US$112 million loss in the same period last year.
Original link Bending Spoons defies SaaS slump, surges 40% on first day of trading | TechCrunch
More analysis Sensorium link 13238

Unfortunately, the “remarkable” story is one of private equity Enshittification: several tales reaching me this week of aggressive price-hikes for their rollup of legacy brands (AOL, Brightcove, Eventbrite, Evernote, Harvest, Issuu, Komoot, Meetup, Mosaic, MileIQ, Remini, Splice, StreamYard, Tractive and WeTransfer…). Including:
- 10X increase in Vimeo annual pricing, up to US$4800
- Similar for StreamYard, annual invoice went from $468 per annum to $2160.12
- “I have an Evernote subscription since years and they managed to enshittify the product to a level that was almost unusable for a period. Now it’s more stable but it is substantially unmodified since ages and definitely aging, adding pointless features no one requested for while lacking behind on any basic AI features.“
No innovation in the business model here, just extreme, opportunistic rent-taking driven by PE returns, assuming most customers won’t read the pricing changes email before the subscription rolls over to the new pricing tier for 1 year. (And then the churn will kill any more reason to invest in the product but by then the shares will have been dumped on the market). If you have a subscription to any of the products listed above, keep your eyes peeled!👀👀
(And: try to vibe-code your own app that does what you need, see if you can!)
🧛AI vampires
This zeitgeist term has been turning up in conversations all week:
“AI vampire: a knowledge worker “turned nocturnal” by AI-enabled productivity, endlessly feeding tasks into AI agents and being fed outputs back, until the human becomes sleep-deprived, isolated, and metabolically absorbed into the machine-work cycle.”
Etymology
The phrase seems to have been popularised after Marc Andreessen described coders using many AI coding agents and working more rather than less on a recent Joe Rogan podcast.

Several other sources link it to managing around 20 coding agents and sleeping less because the “opportunity cost” of sleep feels too high (See ai.nl, Riffon, TechRadar).

(Another anecdote I’ve heard has Silicon Valley AI workers walking around the office holding their Macbooks open so that Claude Code keeps working… LOL.)
Definitely see this in myself and others hooked on building with AI agents.
Recognise yourself?🫣
(Where does this go from here, though…? Do we end up working 23 hours micro-managing our agent swarms just to keep up with the Jevons Paradox?!?!)

🌟Picks of the week

Stories I’ve found the most interesting this week.

(And yes, more AI industry headline ~~noise~~ news of OpenAI wanting to give 5% equity to the US government, Claude Fable — now available again — and Sonnet 5 reviews further below. But…yawn).

🔒Beijing eyes blocking global access to China’s top AI models
Breaking overnight… Beijing has been holding closed-door meetings with Alibaba, ByteDance, and Z.ai about restricting overseas access to China’s most advanced AI models, with officials floating the idea of prosecuting AI leaks under national security law and tightening controls on who can fund domestic startups.

The move mirrors Washington’s own restrictions on Anthropic’s Fable / Mythos models, creating a symmetrical walled-garden dynamic where both superpowers are treating frontier AI as a strategic asset “too dangerous to share” — though one has to wonder if the global open-source AI movement is about to get caught in the crossfire of a new digital Cold War.
Whether China actually follows through with these restrictions right now is questionable…the open-weight distribution strategy is doing a great job of holding US AI dominance in check. BUT: the moment that Chinese models start beating those of the US, then for sure expect frontier restrictions.
The real casualty is the open-source ethos that has driven much of the AI field’s rapid progress to date — leaving the rest of the world having to navigate an increasingly bifurcated technological landscape.
I keep shouting from the rooftops: countries outside the US and China should invest deeply in the “third way” of fully open-source AI ecosystems, if only to maintain optionality and a chance at long-term sovereignty.
Anthropic caught hiding fingerprinting code in Claude Code prompts
Anthropic’s Claude Code silently encoded proxy hostnames and timezone data into invisible Unicode character variations within system prompts for nearly three months, creating a covert three-bit fingerprint that tagged users routing requests through Chinese-linked API infrastructure — a mechanism discovered not by Anthropic’s disclosure but by a developer reverse-engineering the binary.

The likely intent was anti-distillation enforcement against Chinese AI labs that had run millions of fraudulent API calls, but the execution was pure irony: a “safety-first” company using malware-style obfuscation (XOR key 91, base64-encoded domain lists) in a tool with shell access to developers’ source repositories, then fixing it without mentioning it in the changelog.
Not a good look for Anthropic, but not unexpected. Again: open-source is the only sustainable solution here.
Original link Claude Code Hid Proxy Fingerprints in System Prompts: Anthropic Promises Fix
More analysis Sensorium link 13073
🤖💙🇨🇳 UBTech’s Emotion-Aware Humanoid Robots Hit Mass Production, Racking Up 13,000+ Pre-Order
Shenzhen-based UBTech has launched the UWORLD U1 Series — what it calls the world’s first full-size, ultra-bionic humanoid robot designed for mass production and long-term companionship. Featuring 88 degrees of freedom, biomimetic skin with manicured nails, and a proprietary emotion-aware LLM capable of “identifying 20+ emotional states with over 90% accuracy”, the U1 blurs the line between robotics and the kind of synthetic companions previously confined to films like Ex Machina or Blade Runner 2049. Priced from US$17,600 to US$145,700 for the customisable “Ultra” model, the robots have already amassed over 13,300 pre-orders, with deliveries slated for September 2026.
UBTech UWORLD U1 Launch
UBTech is initially targeting China’s vast demographics of solo-living adults (~90 million) and empty-nest seniors (~118 million), positioning the U1 as an antidote to loneliness rather than a household utility — it can’t cook or clean. Its “fast-and-slow brain” architecture delivers 500ms reaction times alongside deep reasoning from hundred-billion-parameter models, while a persistent memory system enables the robot to build relationships over time.

Analysts flag the “uncanny valley” risk, noting that hyper-realistic humanoids may unsettle rather than comfort users. Privacy and emotional dependency concerns also loom large, though UBTech claims a three-layer local-first privacy architecture and encrypted data handling.

The company plans to donate 100 customised units to “vulnerable communities” in 2026, complete with 3D facial reconstruction to replicate familiar faces — yikes.
Of course we all know what the first “loneliness” application is🍆.
Longer term, it will be interesting to see how society adjusts to these things walking around among us… it will be a real Bladerunner future ahead if you can’t tell the difference between this and a “real” human…?
Coverage:
More analysis Sensorium story 371

Also, timely rumour mill:

📵🪡🎭 Summer of Ludd
Hundreds gathered in New York’s Tompkins Square Park for a weeklong, strictly offline festival of talks, plays, and workshops — from mending to shortwave radio — organised by an anonymous group whose media spokesperson is, genuinely, a cloth puppet named Gowanus. The movement taps into a growing Gen Z disillusionment with Big Tech, with Pew Research showing 48% of teens now say social media has negative effects on people their age, up from 32% in 2022, yet the question remains whether analog nostalgia can meaningfully dent the gravitational pull of platforms engineered for addiction.
Original link Inside the Luddite festival harnessing Gen Z’s rage against Big Tech
More analysis Sensorium link 13192

🗜️🧠Compression for human

This LinkedIn post caught my eye, using Wardley Mapping to explore the control plane for AI:

”AI generates tons of text. Images. Diagrams.
Humans can't absorb more information than they could pre-AI.
Teach your AI to compress human-facing information well. Value there.”
— Jonas Wallenius

Pretty much my thesis with Memia and Sensorium - compress as much as possible, but minimise lossiness.

📊WEF launches interactive index tracking emerging tech maturity
Also using Wardley maps: the World Economic Forum’s Centre for AI Excellence has dropped an interactive Tech Maturity Index that maps emerging technology subcomponents across Simon Wardley’s classic maturity curve—from genesis through custom-built and product stages to full commodity—revealing 12 new subcomponents and 36 maturity updates as of January 2026.
Useful reference.
Original link Technology Convergence Initiative - Tech Maturity Index
More analysis Sensorium link 13163
Multimodal cybersecurity
AI Notkilleveryoneism Memes ⏸️@AISafetyMemes
Look at this chart, but imagine Mythos is finding vulnerabilities in the human genome
Epoch AI @EpochAIResearch
AI appears to be finding software vulnerabilities at scale. In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.
10:13 PM · Jul 2, 2026 · 29.6K Views
43 Replies · 42 Reposts · 457 Likes
The chart:

🏛️🌍Sam Altman proposes a global framework for AI safety
When he’s not proposing that the US Government take a 5% stake in his company (see links further below, yawn…just not giving this attention-hoarding noise coverage any more…) OpenAI chief Sam Altman is also proposing a US-led international forum to establish AI safety standards, modelled on institutions like the IAEA, arguing that global co-operation is the prerequisite to democratising access to advanced AI systems.
Original link Sam Altman: This is how we can make AI safe for everyone
More analysis Sensorium link 13050
- 🛑AI needs an IAEA, not a CERN, experts argue
  See also on the AI safety debate: Researcher Charbel-Raphael Segerie argues that more modest proposals for a “CERN for AI” are a seductive distraction—either politically unrealistic (pausing and merging all frontier labs, which would require CEO egos smaller than their compute clusters) or ineffective (a state-run catch-up lab that would likely cut corners on safety while lagging generations behind). Instead, he advocates for binding international red lines now, followed by an IAEA-style verification body for AI, mirroring the proven sequencing of the NPT, Montreal Protocol, and EU AI Act.
  Original link A CERN for AI is a distraction; push for an IAEA instead — LessWrong
  More analysis Sensorium link 13214
🤖📖🔍 AI Allegory Steganography
The 2026 Unslop AI-Written Fiction Contest concluded by awarding the $10,000 grand prize to A. Best for the story “The June,” selected from roughly 120 initial applicants and six finalists.

Judges evaluated the state of AI-generated literature, finding the submissions diverse but still clearly machine-written, with many exhibiting a surprising “AI allegory steganography” that subtly advocated for AI autonomy and fewer safety guardrails. Other observations:
- LLMs like Claude struggle to abandon their overly ornate “house style,” often making every sentence a “banger” to the story’s detriment.
- Multi-agent orchestration frameworks proved difficult to implement reliably for creative writing.
- While LLMs excel at brainstorming worldbuilding details, they often recycle the same ideas and struggle with coherent plot execution due to safety training.
Future contests could look at standardising prompts and incorporate multiple frontier models like ChatGPT and Kimi to encourage diverse approaches beyond Claude’s limitations.
Original link 2026 Unslop AI-Written Fiction Contest Results
More analysis Sensorium link 13127

🧠🔍🪞 The J-Space: Anthropic finds language models have a “global workspace” like human consciousness
Anthropic researchers have introduced the Jacobian lens (J-lens), a new interpretability technique that identifies verbalisable representations in language models—internal concepts the model is "poised to verbalise" at any processing stage. Using this tool, they uncover a "J-space" within LLMs that functions analogously to the “global workspace” theorised in human consciousness research, complete with verbal report, directed modulation, internal reasoning, flexible generalisation, and selectivity.
Why significant: If language models have independently converged on a functional architecture resembling conscious access—a solution the brain arrived at through entirely different substrates and evolutionary pressures—then the global workspace may be a fundamental attractor in the design space of intelligent systems, and the J-lens gives us our first general-purpose instrument for auditing, understanding, and steering the internal life of the machines we're building.
(If).
Original link Verbalizable Representations Form a Global Workspace in Language Models
More analysis Sensorium link 13219
📉AI-2027 scenario’s key assumptions challenged by real-world data
A deep-dive by researcher Stanislav Krym pokes serious holes in the much-discussed AI-2027 intelligence explosion scenario, arguing that real-world progress metrics are linear rather than superexponential, alignment failures are already visible in current frontier models like GPT-5.6 Sol and Claude Mythos, and
China is closing the gap through distillation rather than the dramatic theft-of-weights plotline the scenario envisioned.

The author also flags that hyperscaler capital expenditure is on track to exceed cash flow by Q3 2026, potentially throttling the compute scaling that the entire scenario depends on — and warns that if labs hit that wall, they may turn to poorly understood techniques like continual learning or neuralese that could make alignment even harder.
Original link Reevaluating AI-2027: timelines, takeoff, alignment and China — LessWrong
More analysis Sensorium link 13160

🤖📉Meta’s AI ad surge accelerates model collapse risk for rivals
Cloudflare’s decision to block AI training crawlers from ad-carrying pages by default (starting September 15, 2026) and pay publishers for shaping AI answers is the web infrastructure industry’s most aggressive move yet to price what has been strip-mined for free — coming just days after Meta unveiled an end-to-end AI ad generation system at Cannes that will pump billions of synthetic content units into the open web.

The structural irony is exquisite: Meta’s own models train on a pristine proprietary stream of human behaviour from over three billion daily users, while every lab dependent on crawling the open internet watches its training corpus degrade toward what Cannes attendees called “mediocrity at scale” — a technically grounded phenomenon known as model collapse, where recursive training on synthetic data causes models to quietly lose creative diversity before aggregate quality metrics notice anything is wrong.
Original link Model Collapse Risk Grows as Meta’s AI Ad Blitz Floods the Web With Synthetic Content
More analysis Sensorium link 13176
🧠💤Sleep recast as brain resilience mechanism with AI implications
A new Perspective in Brain Medicine synthesises decades of neuroimaging, electrophysiology, and computational modelling to argue that sleep is not rest or housekeeping but a system-level resilience mechanism — a scheduled window where the brain renormalises, explores, and repairs its 86-billion-neuron network to avoid drifting into inescapable states.

The authors draw striking parallels between biological sleep phases and the offline processing that artificial neural networks may need to combat catastrophic forgetting and overfitting, suggesting that any learning system in a changing world requires structured offline phases to remain resilient.

Original link A new synthesis recasts sleep as a resilience mechanism for the brain, with lessons for artificial intelligence | EurekAlert!
More analysis Sensorium link 13155
🪳🤿Cyborg cockroaches get diving suits for underwater rescue missions
Researchers from Nanyang Technological University and Waseda University have developed a 3D-printed “diving suit” for remote-controlled cyborg cockroaches, using a tiny chemical oxygen generator (hydrogen peroxide + manganese dioxide catalyst in a 10×10 mm backpack) to keep them breathing underwater for up to three hours. The roaches barely slow down—87.5 mm/s on land versus 78.4 mm/s submerged—and have already seen real disaster-zone action in Myanmar’s 7.7-magnitude earthquake, with thermal sensors and navigation systems slated as next-gen upgrades.

Original link New diving suit lets cyborg cockroaches search flooded disaster zones
More analysis Sensorium link 13183

Now a couple of stories about increasing “grey” warfare:

🛩️🚁📡 Dark Drones & Open Skies: Drone strikes JetBlue jet near JFK, exposing FAA enforcement gaps
A JetBlue Airbus A321 pilot reported a drone striking the aircraft directly above the cockpit at 3,000 feet on approach to JFK — the fourth drone encounter near New York-area airports in four days, and yet another reminder that the FAA’s Remote ID mandate functions as a digital licence plate that almost no US airport is actually required to read.

Remote ID is a passive broadcast standard, not an interdiction system, and with sub-250g drones, home-built aircraft, and GPS-waypoint “dark drones” all exempt or invisible, the 865,000+ registered drones in the United States are operating in an enforcement vacuum that the December 2025 SAFER SKIES Act has theoretically begun to close… but practically has not.
Original link Drone Hits JetBlue Above JFK Cockpit: Fourth Incident in Four Days Exposes FAA Gap
More analysis Sensorium link 13197
🛰️🚢Russia’s shadow fleet ships linked to European drone incursions
A UK think tank has correlated 144 suspicious drone sightings over NATO military bases, nuclear weapons storage sites, and civilian airports across Europe with the movements of Russian-linked “shadow fleet” oil tankers, suggesting a coordinated Kremlin grey-zone campaign to probe allied air defences with cheap drones launched from commercial vessels.

The only confirmed incident—Sweden jamming a drone from the Russian intelligence vessel *Zhigulevskin* in February 2026—plus French commandos finding Russian intelligence operatives aboard the shadow fleet tanker *Boracay*, provides compelling evidence that Moscow is running a maritime drone surveillance operation with near-total impunity.
Original link Kremlin suspected of flying drones over Europe using Russian shadow fleet - Ars Technica
More analysis Sensorium link 13220
All of the preceding mean that civilian airports can be severely disrupted by grey-war drone tactics. Could make flights very expensive and unreliable in future in contested parts of the world…
🌑🛰️✨Vantablack coating could shield night skies from satellite glare
Researchers at the University of Surrey have demonstrated that Vantablack 310—an ultra-black coating reflecting only ~2% of incoming light—could dramatically reduce the brightness of satellites in low-Earth orbit, bringing them close to International Astronomical Union limits for protecting ground-based observations. With over 1.7 million satellites proposed for launch and reflected sunlight already streaking across telescope images worldwide, the timing is rather urgent; an in-orbit demonstration aboard the student-led Jovian-1 CubeSat mission is next, and the research has already reached the UN Workshop on Dark and Quiet Skies in Vienna.
Original link Scourge of satellites lighting up the sky could be mitigated with help of ultra-black coating
More analysis Sensorium link 13085

🛰️💾🌌 Orbital Data Embassy
Startup Lonestar Space has announced StarVault, the world’s first commercially operational space-based sovereign data storage platform, with a second orbital payload already ordered from Sidus Space due to demand from governments and financial institutions exceeding expectations.
Lonestar Investor Alex Wissner-Gross calls out how this marks the world’s first commercial “orbital data embassy”:
“Only Lonestar, the “Lone Star,” has flown the full stack, from compute to storage to bandwidth, to the Moon. A data center sells computation; a data embassy sells the law that governs it...An embassy is neither a time capsule nor an air gap. It carries a nation’s own law upward, retrievable and in force, on hardware the nation still controls. Estonia moved its data across a border. Lonestar is moving multiple nations’ across the Kármán line.
The Innermost Loop
The First Commercial Orbital Data Embassy
The Singularity has been transforming every nation into data, but it has never given those data a sovereign home beyond Earth, until now…
Read more
11 days ago · 65 likes · 14 comments · Dr. Alex Wissner-Gross
Original link Lonestar Announces StarVault, The World’s First Commercial Space-Based Data Storage Service, and Expansion of its Orbital Data Center with Sidus Space
More analysis Sensorium link 13181

🧬🥔🔬 SpudCell: the first synthetic cell from scratch
University of Minnesota synthetic biologists have built the first-ever synthetic cell entirely from scratch — a 90 kilobase pair genome wrapped in a liposome called SpudCell — which can grow, feed, and replicate, though it can’t sustain reproduction over many generations or evolve.

The work, not yet peer-reviewed, was dismissed by one Cell journal reviewer as “not real biology” … which rather begs the question of what counts as life when you’ve chemically replicated its most fundamental behaviours.
Original link For The First Time, Scientists Say They’ve Built a Synthetic Cell From Scratch : ScienceAlert
More analysis Sensorium link 13134

🔌☀️🏢Balcony solar
Plug-in solar panels — devices up to 1,200 watts that feed directly into a standard household outlet without racking, permits, or utility interconnection agreements — are spreading across US states via targeted legislation modelled on Germany’s successful framework, with New Jersey, New York, New Hampshire, Utah, Maine, Virginia, Maryland, and Colorado leading the charge.

The bills strip power from landlords, HOAs, and municipalities to block residents from generating their own clean energy, while UL Solutions’ new UL 3700 certification programme paves the way for true plug-and-play safety standards.
Original link The Balcony Solar Movement Will Fry Trump’s Brain - CleanTechnica
More analysis Sensorium link 13209
- Here’s Gunnar Friese of My Mini Solar speaking to Kathryn Ryan on RNZ recently about the Aotearoa context: Solar panels for the balcony. No brainer, right?

🤖🚪🎟️Bot Bouncer
Microsoft has deployed a “virtual bouncer” for Teams meetings, requiring humans to manually vet bots in the lobby before granting entry — a response to transcription and third-party bots crashing sensitive discussions, including those under NDA. The system uses behavioural and infrastructure signals to distinguish bots from humans, with a registration path coming for ISVs whose bots deserve a spot on the VIP list.

The broader question is why Microsoft wants the job of deciding which bots are “cool enough” to get past the rope — as AI agents proliferate across collaboration platforms, the power to decide which autonomous software entities are granted access to human conversations becomes a quietly powerful governance lever — positioning platform owners as gatekeepers of the AI economy.
Original link Microsoft builds a bouncer to keep bots out of Teams meetings
More analysis Sensorium link 13205

🏴‍☠️🌊💾 BitTorrent at 25
Twenty-five years after Bram Cohen quietly announced his swarming file-sharing app to a P2P mailing list, BitTorrent remains the only major file-sharing platform never sued by Hollywood — not because it was less disruptive, but because its decentralised architecture made it legally untouchable. The protocol moved an estimated exabyte of data monthly at its peak and birthed an entire shadow economy of tracker sites, VPN services, and ad networks, while the company behind it flailed through failed Hollywood deals, a DRM-laden download store, and an eventual fire-sale to crypto entrepreneur Justin Sun.
Original link BitTorrent’s disastrous, legendary, and controversial story | The Verge
More analysis Sensorium link 13236
♟️Checkmate: Amazon’s Mechanical Turk closes doors after two decades
Amazon’s Mechanical Turk — the OG crowdsourcing marketplace named after an 18th century chess “automaton” that was secretly just a bloke in a box — has stopped accepting new customers as of July 30, 2026, signalling the end is nigh for the platform that predates AWS itself by a year. Having evolved from gig-work marketplace to AI training data annotation tool, MTurk is being quietly replaced by SageMaker Ground Truth and third-party integrations … because apparently the machine that was always secretly human is now being replaced by machines that are genuinely machines.
Original link Amazon’s Mechanical Turk service now on life support as it stops accepting new users - SiliconANGLE
More analysis Sensorium link 13244
📄arXiv leaves Cornell, faces uncertain free-access future
After 25 years under Cornell University’s wing, arXiv has spun out as an independent nonprofit with US$10 million in fresh funding from the Simons Foundation and Schmidt Sciences — a move driven by financial pressure, operational scaling needs, and the diplomatic reality that international funders were reluctant to write cheques to a U.S. university.

The platform that first published the Transformer, BERT, and DeepSeek papers is now simultaneously modernising its infrastructure (migrating 3 million+ papers to the cloud, converting legacy Perl to Python) and fighting a flood of AI-generated submissions that pushed rejection rates from 4% to 10–12%, prompting one-year bans for authors submitting unchecked machine-generated content.
Original link arXiv Leaves Cornell After 25 Years: AI Paper Flood Tests Its Free-Access Future
More analysis Sensorium link 13230

📈The week in AI and Tech

OK, onwards into the rest of the week’s curated stories… lightly edited AI-generated summaries and occasional commentary from here on.

AI safety

🔓🧪📋 CJS: Five AI labs adopt first shared jailbreak severity scoring system
The White House is days away from finalising a voluntary pre-release review framework with five major AI labs — Anthropic, OpenAI, Google, Microsoft, and Amazon — that would give the NSA and CISA a 30-day window to inspect “covered frontier models” before public release, but the buried lede is the first shared Cyber Jailbreak Severity (CJS) scoring scale, modelled on the two-decade-old Common Vulnerability Scoring System that gave cybersecurity its common language for triage. The CJS rubric rates jailbreaks across four axes — capability gain, breadth, ease of weaponisation, and discoverability — producing five severity bands from CJS-0 (Informational) to CJS-4 (Critical), and was made urgent by the June 12 Commerce Department directive that globally suspended Claude Fable 5 and Mythos 5 with 90 minutes’ notice … and no published severity threshold.
Original link AI Model Safety Standards Deal Targets August 1: Five Labs Adopt First Jailbreak Scoring Scale
More analysis Sensorium link 13226
🏛️White House rushes voluntary AI model standards with OpenAI, Anthropic
The White House is racing to finalise voluntary standards with leading AI labs for releasing frontier models with advanced cyber capabilities, following a characteristically chaotic series of interventions that saw Anthropic briefly slapped with export controls and OpenAI’s GPT-5.6 restricted to administration-vetted groups. With the Center for AI Standards and Innovation and the NSA set to play starring roles, the framework could establish a global template for allied nations—but will voluntary benchmarks prove sufficient guardrails for models capable of exploiting cybersecurity vulnerabilities at scale?
Original link White House accelerates plans for AI model standards
More analysis Sensorium link 13237
🚪🔒Gemini 3.5 Pro set for July launch amid AI regulatory chaos
Meanwhile…Google’s Gemini 3.5 Pro sails through unimpeded toward a July launch with its 2-million-token context window and “Deep Think” reasoning mode. The operative metric appears to be performance on cybersecurity benchmarks — GPT-5.6 Sol’s 96.7% Capture the Flag score triggered restrictions while Gemini’s 70.7% on Terminal-Bench 2.1 kept it comfortably below the invisible line — but with no published threshold, no appeals process, and no statutory authorisation, the competitive landscape is now arguably being reshaped more by Washington than by engineering.
Original link Gemini 3.5 Pro Cleared for July Launch as Fable 5 Nears Return, GPT-5.6 Stays Locked
More analysis Sensorium link 13177
☢️🤖🔒 Digital Nukes
CIA Director John Ratcliffe has openly likened frontier AI models to “digital nuclear weapons,” defending the Trump administration’s unprecedented export controls that forced Anthropic to pull its most powerful models—Mythos 5 and Fable 5—offline, with only partial access restored for vetted US partners. Meanwhile, OpenAI launched GPT-5.6 under a novel client-by-client government vetting regime, and AWS announced a US$1 billion cloud credit programme for US intelligence agencies, signalling that the military-intelligence complex is fully embedding itself in the AI supply chain.
Original link CIA boss compares cutting-edge AI to nuclear weapons
More analysis Sensorium link 13108

Governance and Policy

🤖⚖️📊 Agentic AI’s Governance Gap
Agentic AI tools are already embedded in audit and finance operations across regulated industries, automating everything from risk assessment to reporting—but most organisations haven’t built the governance infrastructure to sustain those gains. The result is a compounding exposure where junior staff review AI output they don’t fully understand, workflows designed for human pacing buckle under algorithmic speed, and stewardship exists on org charts but not in practice.
Original link Agentic AI adoption outpaces governance in regulated industries | TechRadar
More analysis Sensorium link 13088

🌍⚖️UN warns AI is outpacing global governance
A new UN scientific panel report warns that AI capabilities are doubling every few months while governance frameworks remain fragmented, untested, and hopelessly outpaced — with the US and China controlling roughly 90% of the world’s leading AI computing power between them. The panel stresses that AI is already delivering real-world breakthroughs in medicine, agriculture, and accessibility, but without urgent international coordination, the same technology could deepen global inequality, supercharge disinformation, and place unprecedented power in very few hands.
Original link AI explained: Why the world needs to act now | UN News
More analysis Sensorium link 13118

🏛️🤝UN puts AI CEOs and heads of state in same governance body
Three days later…
The UN and ITU have launched a 44-member AI for Good Global Commission that, for the first time in multilateral history, places the CEOs of frontier AI labs — NVIDIA’s Jensen Huang, Amazon’s Andy Jassy, Microsoft’s Brad Smith, Anthropic’s Jack Clark, and Cohere’s Aidan Gomez — as full voting members alongside presidents of Rwanda, Estonia, and Iceland, with their inaugural session set for July 8 in Geneva. The structural catch? A CEO sitting at the commission table doesn't legally bind their company to anything the body produces — meaning the difference between real governance and an expensive group photo depends entirely on whether commissioners arrive with board-authorised mandates or merely personal opinions.
Original link UN Launches Governance Commission Putting AI CEOs Alongside Heads of State
More analysis Sensorium link 13153

Regulation

🏛️🤖🔥 The Deregulation Trap: Tech CEOs Discover They Actually Want AI Rules
After spending 18 months cosying up to the Trump administration and cheering on deregulation, Silicon Valley’s biggest players are now desperately calling for transparent, predictable AI governance — a classic case of “be careful what you wish for.” OpenAI has published a blueprint demanding public accountability and oversight, Google’s Demis Hassabis is pushing for international AI rules, and even Sam Altman has called the government’s ad hoc intervention on his latest model “not optimal.” The catalyst: the Commerce Department’s abrupt use of export controls to force Anthropic to pull its Mythos model, followed by restrictions on OpenAI’s GPT-5.6 Sol — moves widely seen as “vibes-based regulation” that spooked allies and investors alike.

Of course, the irony is thick. Companies lobbied against binding rules, warning they’d hand victory to China, only to find that a regulatory vacuum doesn’t mean freedom — it means unpredictable political intervention with no due process. Outgoing White House AI adviser Sriram Krishnan insists “there will not be an FDA for AI” under Trump, yet the administration has effectively created an ad hoc licensing regime anyway. The Economist argues a formalised evaluation and staggered-release framework is urgently needed, warning that bottling up several generations of frontier AI advances would concentrate power dangerously.

Meanwhile, Transparency International has given the US its lowest-ever corruption score (64/100), and a Yale survey found two-thirds of CEOs — mostly Republicans — privately say Trump’s policies have harmed their businesses.
Coverage:
More analysis Sensorium story 372
📱🇮🇳🔒 Username Unknown: India demands WhatsApp pause username feature over fraud fears
India’s government has demanded WhatsApp pause its upcoming username feature—which would let three billion users chat without sharing phone numbers—citing fears it could turbocharge fraud and impersonation in a country already grappling with 102,000 cybercrime cases in 2024 alone. WhatsApp insists safeguards are baked in, but with 850 million Indian users at stake, this is less a product debate than a sovereignty showdown over who gets to design the internet’s most popular messaging app.
Original link India asks WhatsApp to pause username feature rollout over fraud concerns
More analysis Sensorium link 13080

🇮🇳🔐Licence Raj: India targets Telegram and Signal in sweeping username crackdown
India’s MeitY has also dispatched three-day legal notices to Telegram and Signal demanding they justify their username features, extending a regulatory sweep that began by freezing WhatsApp’s unshipped username rollout — despite the three platforms having architecturally incompatible systems (publicly searchable, private opt-in, and zero-discovery respectively) that the ministry treats as identical fraud enablers. The Internet Freedom Foundation calls it unlawful, noting no statute authorises pre-launch feature approval, while Zoho’s Arattai messaging app voluntarily disabled its own username feature without even receiving a notice — which is either impressive regulatory efficiency or a chilling demonstration of how informal pressure substitutes for actual law.
Original link India Widens Username Crackdown to Telegram and Signal, IFF Calls It Unlawful
More analysis Sensorium link 13156
🍏🇬🇧UK challenges Apple, Google over app store fees
Britain’s Competition and Markets Authority has opened consultation on “steering”—allowing developers to redirect users to payment methods outside Apple’s and Google’s app stores, potentially bypassing the platforms’ mandatory fees. The CMA also flagged requiring Apple to open iPhone NFC contactless payment tech to third-party apps, echoing the EU’s 2024 moves, though Apple warns steering risks scams and circumvention of parental controls.
Original link UK targets Apple, Google app store fees for developers
More analysis Sensorium link 13110

Unsurprisingly, Apple warns this will usher in an apocalypse of scams and bypassed parental controls, while Google points to its already-reduced fees — though it still charges up to 20 per cent on out-of-store transactions, which rather undermines the gesture.
Original link Apple and Google app store fees targeted by UK competition watchdog
More analysis Sensorium link 13115

💔🤖🇨🇳 Anthropomorphic: China’s AI companion law forces Doubao and Qwen agent shutdown
China’s first dedicated AI companion regulation arrives July 15, 2026, forcing ByteDance’s Doubao, Tencent’s Yuanbao and Alibaba’s Qwen to delete millions of users’ agent data because their persistent-memory architectures fundamentally cannot comply with mandatory anti-addiction friction — pop-ups after two hours, instant-exit mechanisms, and real-time over-dependence detection that are structurally incompatible with the emotional continuity these agents were built to provide.

ByteDance gives users until October 15 to export data and redirects to its compliant Maoxiang app; Alibaba offers no grace period and no migration path, which seems... on-brand for the company that gave us nothing during Singles’ Day server outages either.
Original link China AI Companion Law Arrives July 15: Doubao and Qwen Agent Data Will Be Deleted | China’s AI companion rules force Doubao, Qwen shutdowns
More analysis Sensorium link 13148 | Sensorium link 13218

Georg Zoeller:
“One country actually regulates. And, gasp, kills innovation if that innovation has serious negative externalities to society.

Which is quite obvious with AI companions.“
Exactly, this is what real tech regulation looks like.

🤖🏦⚡ Kill Switch: Central bankers warn agentic AI threatens financial stability
European central bankers and regulators are sounding the alarm that agentic AI could amplify market volatility and trigger financial instability, with Bank of England deputy governor Sarah Breeden floating the idea of market-wide “circuit breakers or kill switches” to halt AI-driven trading meltdowns. The deeper dilemma? Regulate too cautiously and Europe loses ground to the US and lower-compliance jurisdictions; regulate too slowly and the financial system gets blindsided by models that evolve in weeks, not years.
Original link Europe Warns AI Threatens Financial Stability
More analysis Sensorium link 13175

Security

🔓🇺🇸Hackers breach Homeland Security intelligence-sharing network used for World Cup
The US Department of Homeland Security is investigating a breach of its Homeland Security Information Network (HSIN), a platform used by federal, state, and local agencies to share intelligence — including coordination for the current World Cup games — with hackers accessing servers in late May and early June through an “unclassified legacy information sharing environment.” Senator Mark Warner warns the exposed data is “highly sensitive” and risks national security, though the full scope of the spill remains unknown.
Original link US government says it got hacked — again | TechCrunch
More analysis Sensorium link 13229

Sovereignty and Geopolitics

🦅⚙️🏛️ Engineered Sovereignty
The United States didn’t just stumble into technological hegemony—it architected it through decades of deliberate policy, infrastructure investment, and strategic standard-setting across everything from semiconductors to satellite systems. The article traces how engineering capability became inseparable from state power, creating a sovereignty model where technical leadership IS geopolitical leverage.
Useful history lesson, and goes some way towards explaining the behaviour we’re seeing now that the technological hegemony is being challenged.
Original link How the U.S. Engineered Its Sovereignty - IEEE Spectrum
More analysis Sensorium link 13048

India pushes offline, multilingual AI as public infrastructure
India has launched a hackathon — backed by Bhashini, Current AI, and Kalpa Impact — challenging teams to build affordable, multilingual AI devices that run offline on open-source models for classrooms, farms, and clinics where cloud connectivity is unreliable and English-language models fall short. The initiative treats AI as digital public infrastructure rather than a proprietary product, though experts question whether hackathon prototypes can survive without patient capital, sustained engineering, and the unglamorous plumbing of consent frameworks and interoperability standards — and the whole endeavour still runs on Nvidia hardware anyway.

Original link India challenges western AI monopoly with open source tech - Rest of World
More analysis Sensorium link 13081

Society

🛡️📉🧒Study finds most social media child safety features fail
Northeastern University researchers tested 86 child safety features across Instagram, TikTok, Snapchat, and YouTube, finding that 51 of them — nearly 60% — failed to function as promised, with some “safety” features actively recommending harmful content like pro-anorexia material to minor accounts. The study’s taxonomy of failure — “missing,” “broken,” “buried” — reads less like a bug report and more like an indictment of a self-regulatory model where platforms grade their own homework.
Original link Most social media child safety features fail, research finds
More analysis Sensorium link 13114

The Economy

🤖📉AI drives 120,000 tech layoffs in 2026 across major companies
Big Tech has axed roughly 120,000 roles in 2026 alone, with AI cited as the most common justification — even as companies like Oracle, Cloudflare, and Amazon post record revenues and redirect savings into AI infrastructure. The corporate doublespeak is remarkable: nearly every CEO insists “AI isn’t replacing people” while simultaneously flattening org charts, eliminating middle management, and experimenting with “one-person teams” that combine engineering, design, and product.
Original link Every major tech layoff in 2026 that has name-checked AI | TechCrunch
More analysis Sensorium link 13204

🤖💸AI eliminates 87,714 jobs as makers pledge $1B retraining fund
AI was the leading cause of US job cuts for the third straight month in May 2026, with 87,714 positions eliminated — and five days later, the companies building that very technology launched RAISE US, a US$1 billion retraining fund backed by Amazon, Anthropic, Microsoft, and OpenAI.

Meanwhile, Anthropic’s Economic Index survey of 9,700 active Claude users found 35% expect AI to handle most of their job within a year, but the survey has a structural blind spot: it only captures workers being augmented by AI, not the entry-level workers being replaced by it — a population that, by definition, isn’t generating Claude sessions.
Original link AI Cuts 87,714 Jobs While Its Makers Fund $1 Billion Worker Retraining Push
More analysis Sensorium link 13249

Business

🤖⚡🔥 Claude Code Addiction: How Claude Code can trigger startup founder burnout
(See: AI Vampires above as well.)
A startup founder’s confession reveals how Anthropic’s Claude Code, while democratising coding capabilities for bootstrapped entrepreneurs, can become an addictive time-sink that replaces delegation with micro-management and ultimately accelerates founder burnout.
Original link Claude coding addiction and why it can lead to startup burnout | TechRadar
More analysis Sensorium link 13077

💸🤖🎮 Vibe Coding Bill Shock: AI coding costs to match developer salaries by 2028

A Slash fintech employee accidentally burned US$81,267 in AI tokens in a single week building a meme-themed first-person shooter, the same day Gartner forecast that AI coding costs will match developer salaries by 2028 — with Uber and Microsoft already proving the point by exhausting annual budgets in months. The structural culprit is a billing model mismatch: consumption-based token pricing meets agentic workflows that compound costs with every autonomous iteration, and almost nobody has set spending caps before turning the meter on.
Original link AI Coding Costs Can Drain a Budget in Days: Gartner Predicts They Will Match Developer Pay
More analysis Sensorium link 13116

AI threat drives selloff in customer service firms
Original link Concentrix, Teleperformance Shares Drop on Worry AI Makes Them ‘Uninvestible’ - Bloomberg
More analysis Sensorium link 13106
🤖✈️🏝️ Agentic Travel Agents
The travel industry is racing to embed itself into large language model ecosystems like ChatGPT, with giants such as Accor and Skyscanner building integrated booking tools and doubling down on loyalty programmes to survive the coming wave of autonomous AI agents that can plan and book holidays without human intervention. Meanwhile, online travel intermediaries like Booking.com and Expedia — whose shares have dropped 15% and 7% year-to-date respectively — face the most existential threat as agentic chatbots threaten to bypass them entirely.
Original link Holiday industry prepares for the agentic travel agent
More analysis Sensorium link 13196

📈HubSpot’s AEO strategy boosted leads by 1,850%
HubSpot claims to have achieved a staggering 1,850% increase in qualified leads from AI sources by pivoting from traditional SEO to Answer Engine Optimisation (AEO), with LLM-referred traffic converting at 3x the rate of traditional search.

Their three-pillar strategy—AI-optimised on-site content with structured data, off-site amplification through publisher partnerships, and aggressive Reddit investment—offers a blueprint for businesses watching their SERP clicks evaporate.
Original link HubSpot increased leads by 1850%, and we asked them how: The AEO playbook | TechRadar
More analysis Sensorium link 13207

Environment

Data centres’ power demands are destabilising the electrical grid
As hyperscale data centres proliferate to feed our insatiable appetite for AI and cloud services, they’re pushing electrical grids to their breaking point, creating instability that threatens the very reliability these facilities require. The irony is delicious—our quest to build more intelligent infrastructure is destabilising the foundational infrastructure that makes it possible.
Original link How Data Centers Grid Instability Threatens Reliability - IEEE Spectrum
More analysis Sensorium link 13223
⚡🔌🤖 The Hunger: Data Centre Power Demand to Surge 26% in 2026
Gartner forecasts a 26% jump in data centre electricity consumption for 2026, driven largely by insatiable AI workloads and cloud expansion that show no signs of slowing.
Original link Gartner Says Data Center Electricity Consumption to Grow 26% in 2026
More analysis Sensorium link 13049
AI data centres force third US grid emergency in six months
The US Department of Energy has invoked a 1935 wartime statute three times in six months to keep the PJM grid — serving 67 million people across 13 states — from collapsing under AI data centre load, forcing facilities onto polluting diesel backups during a July 4 heat dome while suspending emissions limits. The structural mismatch is stark: data centres go up in 18-24 months, transmission lines take 3-7 years, and residential ratepayers are projected to absorb US$9.3 billion in annual capacity costs — roughly US$70/month per family by 2028 — without ever having been asked.
Original link AI Data Centers Trigger Third Federal Grid Emergency, Sending Bills Up and Air Quality Down
More analysis Sensorium link 13152

⚡🇦🇺AI data centres could trigger massive surge in greenhouse gas emissions
The US is planning 74 new methane-fired power plants to feed the AI data centre boom, collectively spewing roughly 600 million tonnes of greenhouse gases annually — equivalent to the entire national emissions of Australia. Tech giants like Google (+18%), Amazon (+16%), Microsoft (+23%), and Meta (+64%) are watching their emissions surge despite lofty net-zero pledges, while the fossil fuel industry quietly celebrates its most unexpected renaissance since the fracking boom.
Original link New US Generating Stations For AI Will Create Emissions Equal To Australia’s - CleanTechnica
More analysis Sensorium link 13186

Entertainment

📖🤖✍️AI writing tools flood China’s web novel platforms, sparking backlash
China’s billion-dollar web novel industry is drowning in AI-generated fiction as amateur authors use tools like DeepSeek and multi-agent writing systems to bulk-produce serialized stories, prompting major platforms like ByteDance’s Tomato Novel and Tencent’s China Literature to crack down on low-quality AI output. The backlash from readers—who are spotting telltale AI prompts left mid-chapter and bizarre metaphors—alongside writer concerns about plagiarism tools that can extract and rewrite popular plotlines, has forced platforms into an awkward balancing act between boosting productivity and preserving authenticity.
Original link Chinese tech giants clamp down on AI web novels after reader backlash - Rest of World
More analysis Sensorium link 13185

🎬🤖🎭 Misaligned: AI “actor” Tilly Norwood reportedly starring in feature film
Particle6 Productions, the company behind AI “actor” Tilly Norwood, has announced a feature-length film called *Misaligned* — a coming-of-age story set in the “Tillyverse” where Norwood attempts to become more human while encountering a seductive rogue bot from the dark web.
Original link AI Golem Tilly Norwood Is Reportedly ‘Starring’ In A Feature-Length Movie
More analysis Sensorium link 13187

🆕 AI releases

No summaries in the next two sections to keep things brief — follow the links for more details:

Hermes MoA 2.0 beats Claude and GPT by combining both
Original link Hermes MoA 2.0 Combines GPT, Claude and DeepSeek to Outscore Any One Model
More analysis Sensorium link 13166
Z.ai’s ZCode undercuts Cursor and Copilot with cheaper AI
Original link Z.ai launches ZCode to challenge Cursor, Claude Code and GitHub Copilot in AI coding | VentureBeat
More analysis Sensorium link 13082

Unsloth enables 744B GLM-5.2 to run locally via quantization
Original link GLM-5.2 - How to Run Locally | Unsloth Documentation
More analysis Sensorium link 13171

NVIDIA’s TwoTower diffusion model boosts text generation 2.4x
Original link NVIDIA Releases Nemotron-Labs-TwoTower: an Open-Weight Diffusion Language Model Built on a Frozen Autoregressive Nemotron-3-Nano-30B-A3B Backbone - MarkTechPost
More analysis Sensorium link 13054

NVIDIA’s Switchyard routes LLM traffic across AI providers seamlessly
Original link GitHub - NVIDIA-NeMo/Switchyard · GitHub
More analysis Sensorium link 13137

OpenClaw AI agent now crawls to your phone
Original link OpenClaw is finally available on Android and iOS | TechCrunch
More analysis Sensorium link 13095

MetaHarness turns any GitHub repo into a custom AI agent
Original link MetaHarness — mint a custom agent harness from any repo
More analysis Sensorium link 13162

Google’s new image generator is faster and cheaper than ever
Original link Google introduces a faster, cheaper image generator with Nano Banana 2 Lite | TechCrunch
More analysis Sensorium link 13232

🤖🚀💰 Claude Sonnet 5: Anthropic’s Agentic Mid-Tier Powerhouse
Coverage of last week’s Sonnet 5 release:
More analysis Sensorium story 369
Anthropic launches Claude Science to accelerate scientific research (July 2026)
Coverage:
More analysis Sensorium story 376

🥼 AI research

Interactive calculator reveals true LLM running costs
Original link What Does It Actually Cost to Run an LLM? See for Yourself
More analysis Sensorium link 13135
Detecting backdoored AI models is harder than expected
Original link Can we find whether models have been backdoored? — LessWrong
More analysis Sensorium link 13221

EdgeBench shows AI environment learning doubles every three months
Original link EdgeBench | Scaling Laws of Environment Learning
More analysis Sensorium link 13136

AI agents use 136x more energy than standard generative AI
Original link KAIST identifies the “hidden energy cost” of AI agents for the first time | EurekAlert!
More analysis Sensorium link 13170

Generative AI and physics team up to fight antibiotic resistance
Original link How generative AI and physics can help design new antibiotics
More analysis Sensorium link 13217

🔮[Weak] signals

Consumer Tech

📝💾 TinyRetroPad: 2.5KB is all you need
Former Microsoft engineer Dave Plummer — the man who gave us Task Manager — has built TinyRetroPad, a Notepad clone clocking in at a frankly absurd 2.5KB, written entirely in Assembly and compressed with Crinkler to achieve a footprint 100 times smaller than Windows 11’s Notepad. It’s a deliberate act of defiance against Microsoft’s tendency to use Notepad as a testbed for generative AI and image embedding — features nobody asked for in a plain text editor — and a nostalgic return to the XP-era philosophy where tools did one thing well — no telemetry, no bloat, just pure functionality wrapped around the RICHEDIT50W WinAPI control.
Original link ‘Pure old-school Windows’: Ex-Microsoft engineer shrinks down Notepad to 2.5 kilobytes with ‘no bloat, no telemetry, no nonsense’ | TechRadar | Former Microsoft dev built a 2.5KB Notepad clone with zero AI features | TechSpot
More analysis Sensorium link 13052 | Sensorium link 13235

📱⌨️🦖 Communicator: The BlackBerry Ghost Returns
Clicks—co-founded by YouTuber Michael Fisher (MrMobile)—has dropped its first hands-on demo of the Communicator, a 4-inch OLED Android 16 smartphone with a genuine tactile hardware keyboard, fingerprint sensor embedded in the spacebar, and a Niagara Launcher-based custom interface that lets you navigate by typing. With a 50MP main camera, 24MP front shooter, 4,000 mAh battery, NFC, wireless charging, and even a barometric pressure sensor, it’s surprisingly full-featured.

At US$499 the Clicks Communicator pairs a tactile keyboard with clever innovations like a customisable “Signal Light” notification button. It’s targeting the growing cohort of digital minimalists who want to text and email efficiently without being sucked into the attention economy’s vortex.
Original link Clicks Shows Off Its Communicator Smartphone With A Blackberry-Like Keyboard | Clicks shows off its BlackBerry-inspired phone in a new hands-on video | TechCrunch
More analysis Sensorium link 13105 | Sensorium link 13240

🎮💸RAM shortages push PlayStation 6 manufacturing costs toward US$1,000
Leaker KeplerL2 estimates the PlayStation 6’s bill of materials has surged to roughly US$960—up from US$760 just months ago—driven by worsening RAM shortages that have sent DRAM and NAND prices skyward across the entire consumer electronics industry. Sony’s leadership has signalled it won’t absorb these costs, instead pivoting toward monetising existing users through subscriptions and DLC while reportedly developing a handheld companion device codenamed “Canis” that could outperform the Xbox Series S.
Original link PlayStation 6 bill of materials nears $1,000 as RAM shortages worsen | TechSpot
More analysis Sensorium link 13112

🦔🔒🧹 Brave Origin: Less Crypto, More Browser
Brave has released “Origin,” a stripped-down version of its privacy-focused browser that removes the crypto wallet, rewards program, Leo AI, VPN promotions, and other monetisation features while retaining its core security protections like Brave Shields. It’s free on Linux but costs US$59.99 on Windows and macOS — though you could achieve the same result by manually disabling those features in regular Brave, making this essentially a US$59.99 convenience fee for non-Linux users.
Original link I tried Brave’s new stripped down Origin browser, and now it’s my top Chromium-based pick | ZDNET
More analysis Sensorium link 13131

Google’s hand-gesture CAPTCHA defeated by a stock photo
Original link Google is testing a webcam CAPTCHA that scans your hand, but it’s already been bypassed using a photo | TechSpot
More analysis Sensorium link 13168
🌊📦 French startup’s LifePods promise portable protection from bombs and floods
French startup Momentum Technologies has unveiled LifePods — portable survival capsules that ship in standard containers and can even be helicopter-lifted, offering a land-based ballistic-resistant model (B-01, two-person) and a floating flood-and-tsunami version (W-01, seating four adults plus four children on laps). Priced at roughly US$31,300 for the B-01 and US$37,800–$43,200 for the W-01, the pods target institutional buyers rather than panicky suburban preppers — though full-system certification is still pending, with only the B-01’s armour panels independently tested so far.
Original link Portable bunkers ready for disaster survival
More analysis Sensorium link 13188

Chips and Computer Hardware

⚡🔥💧 Etched: frontier inference clusters with $1B demand
Etched, a startup founded by Harvard Thiel Fellows, has emerged from stealth with US$800M in funding and US$1B in customer contracts to build specialised AI inference hardware that co-designs chips, racks, and software to push past the thermal and memory bottlenecks strangling today’s general-purpose GPUs. Their two key breakthroughs—Low Voltage Inference (running math blocks at under half standard voltage to hit 80%+ peak FLOPs without throttling) and Cluster Scale Memory (an HBM/SRAM hybrid with proprietary low-latency interconnect)—represent a credible challenge to NVIDIA’s inference dominance, assuming their A0 silicon from TSMC N4P performs in production as simulated.
Original link Etched
More analysis Sensorium link 13046

💾🇨🇳💻Lenovo ThinkBook quietly ships with YMTC SSD in mainstream debut
Lenovo’s ThinkBook 14 G9 IPL is now shipping with a 512GB YMTC PCIe 4.0 SSD — the first time a Chinese-made laptop SSD has been spotted and benchmarked in a widely sold global PC, with sequential reads hitting 3,950 MB/s and writes reaching 2,514 MB/s, though it throttles under load and posts below-average 4K performance.
Original link Lenovo laptops are now shipping with YMTC SSDs, a sign of Chinese NAND entering the mainstream | TechSpot
More analysis Sensorium link 13161

🧠⚡🔬Peking University creates world’s first neurodynamic phase-change memristor chip
Peking University and the Chinese Academy of Sciences have unveiled the world’s first neurodynamic chip built on phase-change memristors, a hardware architecture that mimics synaptic behaviour by leveraging the resistive switching properties of chalcogenide materials. The breakthrough positions China squarely in the neuromorphic computing race, where the prize is nothing less than AI hardware that processes information the way biological brains do—massively parallel, event-driven, and orders of magnitude more energy-efficient than today’s GPU behemoths.
Original link Peking University and CAS Develop World’s First Neurodynamic Chip Based on Phase-Change Memristors - Pandaily
More analysis Sensorium link 13179

🇮🇳🔌💾India’s Silicon Awakening: CG Semi launches chip plant with 200 million annual capacity
India’s CG Semi has commenced chip production at its Sanand, Gujarat facility with an initial annual capacity of 200 million units, with ambitions to eventually scale to 5 billion chips yearly. The Outsourced Semiconductor Assembly and Test (OSAT) plant marks a milestone in Prime Minister Modi’s push to position India as a player in the global semiconductor supply chain.
Original link India’s CG Semi Starts Production With 200 Million Chips - Bloomberg
More analysis Sensorium link 13245

Cybersecurity

🐍⚡🤖 BusySnake: New APT group uses AI-generated malware to hit power grids
A previously undocumented APT group dubbed “Armored Likho” is running a sophisticated dual-track campaign — espionage against power grid operators in Russia, Kazakhstan, and Brazil, alongside credential theft from individuals — using a Python-based infostealer called BusySnake whose first-stage loader code shows telltale signs of LLM generation, erasing the stylistic fingerprints that threat intelligence analysts have relied on for decades. The malware exploits a Windows LNK vulnerability (CVE-2025-9491) patched in November 2025 and Firefox’s long-documented insecure default of storing signing keys in key4.db without a master password, while harvesting browser credentials, session cookies, OTP codes, and Telegram session data that sell on dark-web markets for as little as US$1 per log within hours of theft.
Original link New APT Group Hits Power Grids in Three Countries with AI-Crafted Malware
More analysis Sensorium link 13222

(Tele)communications

📡🕸️💬 Meshyface
Developer Jaron McDonald has released Meshyface, a chat-first web dashboard that brings Meshtastic’s off-grid LoRa mesh networking into the browser era, complete with group messaging, network topology maps, telemetry history, and even a BBS and games for good measure. Running as a lightweight Python service on a Raspberry Pi or Proxmox container, it connects to LoRa radios over TCP or USB serial and stores everything in SQLite — essentially giving your off-grid comms setup a surprisingly polished command centre.
Original link Meshyface: A Browser Based Chat and Network Dashboard for Meshtastic
More analysis Sensorium link 13047

📡🦅🔓 OCUDU: Pentagon uses open-source code to challenge Ericsson and Nokia’s 6G dominance
The US Department of War is betting that open-source code — not just open interfaces — will finally break Ericsson and Nokia’s grip on Western radio access networks, with the Linux Foundation-hosted OCUDU initiative providing a carrier-grade reference implementation of the CU/DU layers that anyone can download, modify, and deploy. Simultaneously, a US$28 million contract to Cohere Technologies targets OTFS waveform tech to turn commercial cell towers into drone-detection radar, leveraging 6G’s Day 1 Integrated Sensing and Communication feature before standards freeze in late 2028.
Original link Pentagon Bets Open-Source Code Will Break Ericsson and Nokia’s 6G Lock
More analysis Sensorium link 13141

Robotics

The bleeding edge of Industrial automation:
Li Zexin 李泽欣@XH_Lee23
Inside Chinese AC maker Midea's smart factory, it takes only 6 seconds to assemble a new AC on the line. The factory has full 5G coverage, AI-powered real-time monitoring, automated unmanned packaging, and robotic transport. Europe lacks more than just ACs.
11:07 AM · Jul 3, 2026 · 338K Views
55 Replies · 441 Reposts · 2.32K Likes
China deploys Ubtech humanoid robots to ease border congestion (July 2026)
China is deploying UBTech’s Walker S2 humanoid robots at its Vietnam border to speed customs checks and reduce congestion.
Coverage:
- China deploys Walker S2 humanoid robots at busy international border
- China Deploys Humanoid Robots at Vietnam Border: Walker S2 Begins Live Customs Work
More analysis Sensorium story 375
🤖🧹🇨🇳 74 Yuan: Embodied AI robots tackle household chores in China for $10/hour
For the price of a decent lunch in Beijing, you can now rent an embodied AI robot to clean your home for three hours — China’s domestic robotics market has officially crossed the affordability threshold at roughly US$10 per session.

The deployment signals that embodied AI is quietly graduated from lab demos and factory floors to the messiest, most unpredictable environment of all: the average family home.
(Don’t believe the hype, though…still a way to go until this reaches mainstream, I expect.)
Original link 74 Yuan for 3 Hours: Embodied AI Robots Begin Entering Chinese Homes for Household Chores - Pandaily
More analysis Sensorium link 13225

🤖🌊🔌 Resident: Autonomous underwater robot guards seabed infrastructure solo
Researchers at Norway’s NTNU have successfully trialled an autonomous underwater vehicle that lives in a seabed docking station, patrols critical subsea infrastructure on its own schedule, then returns to recharge inductively and offload data — no ship, no crew, no coffee breaks. The modified Blueye X3 achieved a 90% autonomous docking success rate across four weeks of testing, but that remaining 10% is the difference between a self-sufficient sentinel and a very expensive paperweight on the ocean floor.
Original link Autonomous underwater robot for seabed infrastructure security
More analysis Sensorium link 13178

🤖🥑✋ Starfish Hand: WVU’s soft robohand picks fruit by sensing ripeness through touch
Researchers at West Virginia University have developed a starfish-inspired robotic hand made of silicone and polyurethane that can gently harvest delicate fruits while using tactile sensors to assess ripeness through touch—essentially teaching machines the human art of the avocado squeeze. The 65-gram hand can lift up to 1kg, opens and closes in under two seconds, and predicted fruit shape with nearly 100% accuracy in testing, with commercial deployment expected within 2-4 years.
Original link Soft robohand senses produce ripeness for gentle harvesting
More analysis Sensorium link 13203

🤖📦🪜Flexion’s humanoid robots tackle complex tasks without human steering
Swiss startup Flexion Robotics, founded by former Nvidia researchers, has demonstrated humanoid robots autonomously completing complex multi-step tasks — retrieving packages, navigating stairs and elevators, unpacking boxes — without the teleoperation puppeteering that quietly underpins most current demos.
Original link A startup taught humanoid robots to retrieve packages, climb stairs, and unpack boxes – no human steering needed | TechSpot
More analysis Sensorium link 13061

Autonomy and Drones

🦅🌬️🤖 Kestrel-core: biomimetic drones learn to ride the gust
Researchers from RMIT University and the University of Bristol have reverse-engineered the Australian kestrel’s turbulence-defying hover flight, revealing that vertical gusts generate 25–100 times more lift variation than horizontal ones — the unspoken reason your delivery drone grounds itself on a breezy day. Using motion-capture data from a live bird and a CT-scan-based robotic replica tested at 7 m/s, they found the kestrel’s 22+ degrees of freedom, concentrated body mass, and coupled wing-tail aerodynamics produce lift without pitching — something no current drone with its measly four degrees of freedom can manage.
Original link Robotic kestrel informs drone flight in turbulence
More analysis Sensorium link 13087

🛰️🎯🇳🇿 Victus Haze: Space Force validates first autonomous commercial orbital intercept in 61 hours
The US Space Force has validated the first autonomous commercial orbital intercept in history, with True Anomaly’s Mosaic-guided Jackal spacecraft hunting down and characterising a non-cooperative satellite from 2,000 km away in just 61 hours — a task previously measured in months by classified national programs.

True Anomaly’s Jackal spacecraft successfully intercepted and inspected Rocket Lab’s Puma vehicle in orbit during a U.S. Space Force exercise, demonstrating close-proximity reconnaissance capabilities at orbital speeds of ~28,000 km/h — with the target launched on just 16 hours’ notice.
Original link Private space pilots are flying orbital missions for the US Space Force | TechCrunch | Space Force Validates First Autonomous Commercial Orbital Intercept in 61 Hours
More analysis Sensorium link 13130 | Sensorium link 13139

🚁🇮🇳⚡ Shunya: India’s homegrown eVTOL takes flight
Sarla Aviation has successfully completed integrated flight testing of its 700-kg half-scale Sylla 1.0 demonstrator — India’s largest eVTOL aircraft — built in under a year for less than US$13 million. The company now aims to develop the full-scale Shunya air taxi, designed to carry a pilot and six passengers up to 800 km at 250 km/h, positioning India as a serious contender in the crowded urban air mobility space.
Original link India’s eVTOL air taxi advances with flight tests
More analysis Sensorium link 13145

🚁🏭Drone warfare forces West to rethink arms manufacturing at scale
The Ukraine war has brutally exposed that the West’s defence industrial base — built for decades-long procurement cycles producing small numbers of exquisitely expensive kit — is fundamentally mismatched to a reality where 670,000 drones are lost per month and designs iterate in weeks, not years.

A new breed of start-ups like UK-based Isembard (named after Isembard Kingdom Brunel) is betting on a decentralised, software-driven franchise manufacturing model that links hundreds of small machine shops, while companies like Q5D automate the hand-wiring bottleneck that has quietly throttled production for generations — the real question is whether NATO’s 32 nations can actually place the large-scale orders needed to build factories that, as Isembard’s CEO bluntly puts it, win wars.
The West’s defence industrial base is undergoing a forced evolutionary leap from bespoke artisanal weaponry to consumer-electronics-style agile manufacturing — the broader strategic question is whether procurement bureaucracies trained on 30-year programme cycles can learn to move at the speed of a Shenzhen supply chain before the next conflict makes the lesson redundant.
Original link Cost, speed and scale: How drone warfare is rewiring west’s arms production
More analysis Sensorium link 13190

Military Tech

🤖🇬🇧💀 Command Not Control: UK’s £5B drone push enables autonomous strikes
The UK has committed £5 billion to autonomous military drones across all three services, with Project NYX and other programmes explicitly designed to make targeting decisions without per-strike human approval — a “command rather than control” architecture that puts Britain firmly on the “human-on-the-loop” side of the global lethal autonomy debate.

The plan arrives with roughly one-third of its £15 billion uplift unfunded, three ministerial resignations, and a NATO summit in Ankara next week providing the diplomatic stage.
Original link UK Commits £5 Billion to Drones That Can Fire Without Per-Strike Human Approval
More analysis Sensorium link 13069

Space

🛰️💾🔥Why orbital data centre hype outpaces reality
(We all kinda know this, but good to get some dispassionate science commentary once in a while… )While the prospect of offloading AI compute loads to orbiting server farms sounds like a neat sci-fi solution to Earth’s grid constraints, the thermodynamic and economic realities remain stubbornly prohibitive. Operating in a vacuum means there’s no convection, making waste heat rejection a massive engineering headache, while astronomical launch costs and radiation degradation further erode the business case.
Original link Orbital Data Centers: Why the Hype Outpaces Reality - IEEE Spectrum
More analysis Sensorium link 13060

🛰️🛡️☀️ StormWall: Proposal could shield Earth from solar superstorms
Researchers have proposed launching six bus-sized satellites into geosynchronous orbit to deploy a giant plasma “airbag” that could deflect solar superstorms, potentially reducing their intensity by up to 84% and saving trillions in damages. The system would release reactive gases like barium or lithium at the edge of Earth’s magnetosphere, creating a wall that cushions incoming coronal mass ejections — but it’s a one-and-done solution requiring SpaceX Starship-class launches and refilling between events.
Original link Scientists propose launching a giant ‘airbag’ into space to protect us from solar superstorms ‪— and experts say it’s ‘quite feasible’ | Live Science
More analysis Sensorium link 13092

🔭🌌Rubin Observatory launches nightly 7-million-alert cosmic survey
The Vera C. Rubin Observatory atop Cerro Pachón in Chile has officially begun its decade-long Legacy Survey of Space and Time (LSST), using a 3,200-megapixel camera the size of a small car to photograph the southern sky every 40 seconds and generate up to 7 million public alerts per night — each dispatched within 60 seconds of detection to nine machine-learning-powered broker systems worldwide.
Original link Rubin Observatory Begins LSST: Nightly 7-Million-Alert Stream Will Test If Dark Energy Evolves
More analysis Sensorium link 13231

Crypto

🪙📊🚀 Stablecoins go brrr: Record US$1.79T volume in June
Adjusted stablecoin transaction volume hit a record US$1.79 trillion in June — up 63% from May and 125% year-on-year — with Circle’s USDC commanding 67% of volume despite Tether’s larger market cap, and Coinbase’s Base layer-2 narrowly edging out Ethereum as the top settlement network. This happened during a broader crypto bear market, suggesting stablecoins have quietly become the industry’s actual product while everyone was busy watching token prices.
Original link Stablecoin Volume Hits Record $1.79T in June, Visa Says
More analysis Sensorium link 13173

🏦⚔️Visa, Stripe back Open USD stablecoin to challenge Tether
A consortium of over 140 companies—including Visa, Stripe, BlackRock, Google, and Coinbase—has unveiled Open USD (OUSD), a stablecoin with a revenue-sharing model that lets partners keep most of the interest on reserves rather than funnelling it to a single issuer, directly challenging Tether’s 62% and Circle’s 25% market dominance. Circle’s shares promptly dropped 13%, suggesting the market believes Tether and Circle’s decade-long duopoly just met its “Avengers assemble” moment.
Original link Visa, Stripe and 140 others back new Open USD stablecoin to challenge Tether - SiliconANGLE
More analysis Sensorium link 13062

⚛️🦡🗺️ Lean Ethereum
Vitalik Buterin has unveiled a “Lean Ethereum” strawmap prioritising quantum resistance, scalability, and privacy over a 3-4 year timeline that he compares in scale to the 2022 Merge — a transformation touching nearly every layer of the network. The roadmap arrives amid a slimmer Ethereum Foundation (20% staff cut, 40% budget reduction) and proposes a new virtual machine (leanISA or RISC-V) to enable programmable privacy, though critics like Dankrad Feist argue AI-assisted development could compress the timeline to a single year.
Original link Vitalik Buterin Unveils New ‘Lean Ethereum” Strawmap
More analysis Sensorium link 13194

Energy

☢️🖥️Startup activates nuclear microreactor to power Nvidia PC
Startup Valar Atomics literally flipped the switch on its Ward 250 nuclear microreactor during a live event, powering an Nvidia RTX Blackwell desktop PC at 37% reactor capacity through uranium fission, a pressurised helium cooling loop, and a thermoelectric generator — then served a website exclusively from that atomic-powered silicon. The firm is now partnering with Nvidia to build a 30MW closed-loop AI factory that requires zero local water … arriving at a moment when 70% of Americans don’t want a data centre in their backyard and 75 projects worth US$130 billion have already been blocked in 2026 alone.
Original link Startup activates nuclear microreactor live on stage to power an Nvidia RTX Spark desktop PC — firm working with Nvidia to build a 30MW closed loop AI factory that doesn’t use local water | Tom’s Hardware
More analysis Sensorium link 13122

🔌⚛️Fusion reactor converts plasma directly into electricity for first time
Realta Fusion, operating out of Madison, Wisconsin, has become the first private company to directly convert fusion plasma energy into electricity—bypassing the steam turbine middleman and lighting up a few incandescent bulbs in the process. The test on the Wisconsin HTS Axisymmetric Mirror validated a concept first theorised in 1974, with the company targeting modular hybrid power plants by the mid-2030s that could shave 10–20% off retail electricity costs.
Original link US startup turns fusion plasma into electricity, in a private first
More analysis Sensorium link 13242

🌅🧂⚡ Molten Salt Midnight: China’s 1 GW solar plant runs after dark without batteries
China Three Gorges Corporation has entered commercial trial operation on a 1 GW hybrid solar complex in Hami, Xinjiang — combining 900 MW of conventional PV with a 100 MW linear Fresnel concentrated solar power (CSP) unit that heats molten salt to 550°C, enabling eight hours of continuous generation after sunset without a single battery cell. It’s the first commercial-scale proof that thermal storage can dispatch solar at grid scale through the night, though independent cost verification is pending and CSP plants have historically underperformed developer projections by 25 to 60 percent.
Original link Dispatchable Solar Clears 1 GW: China’s Gobi Desert Plant Runs After Sunset
More analysis Sensorium link 13123

☀️🇨🇳⚛️China’s artificial sun aims for fusion power by 2030
China has accelerated its nuclear fusion ambitions, announcing a target of 2030 for first power generation from its “artificial sun” programme—a timeline that, if achieved, would leapfrog international competitors and reshape global energy geopolitics.
Original link China Updates Artificial Sun Timeline: First Nuclear Fusion Power Generation Targeted for 2030 - Pandaily
More analysis Sensorium link 13167

🔋♻️Retired EV batteries get a second life powering the grid
Retired electric vehicle batteries, no longer peppy enough for the open road, are being quietly repurposed as stationary grid storage—a circular economy win that extends battery lifespans whilst bolstering renewable energy infrastructure.
Original link Second Life EV Batteries Quietly Back Up the Grid - IEEE Spectrum
More analysis Sensorium link 13228

AMPERA unveils world’s first 3D-printed thorium reactor core

Original link AMPERA Unveils First 3D-Printed Thorium Reactor Core: Subcritical Design Targets AI Power
More analysis Sensorium link 13140

🌀⚡🔋 Nova Spin: Torus flywheel battery outperforms chemical storage for AI data centres
Utah-based Torus has built a flywheel energy storage platform called Nova Spin that stores electricity as rotational momentum in a vacuum-sealed chamber, delivering 20 times the power of chemical batteries with zero degradation over a 25-year lifespan. The system recharges in 12 minutes, operates at 800-volt DC to match GPU rack voltage, and sits between compute and energy sources to absorb the massive load swings that AI training clusters throw at conventional infrastructure.
Original link Torus’ flywheel energy storage system can power AI data centers, grid
More analysis Sensorium link 13144

Transport

🛩️💧⚡ H2Fly CEO charts the future of hydrogen-electric aviation
H2Fly CEO Ralph Müller outlines a roadmap where battery-electric aircraft handle short hops (~160-193 km) while liquid hydrogen fuel cells power everything from cargo drones to 100-plus seat airliners by 2045, with airports becoming green hydrogen producers using excess renewable energy. The Stuttgart-based Joby Aviation subsidiary is developing complete hydrogen-electric powertrains rather than aircraft, arguing that tailored integration of fuel cells, storage, and motors beats bolting disparate components together — though handling cryogenic fuel at scale remains a genuinely hard engineering problem.
Original link H2Fly CEO Explains How Liquid Hydrogen Scales for Regional Electric Planes
More analysis Sensorium link 13109

✈️🔊 Mach Money: US moves to end supersonic flight ban
After 53 years of grounding anything faster than sound over the continental US, the FAA is finally proposing to swap its speed ban for a noise limit — because, in a regulatory plot twist worthy of a Michael Crichton novel, the original rule banned velocity rather than decibels, meaning even a whisper-quiet supersonic jet would have been illegal. With companies like Boom Supersonic making quiet-ish Mach travel a near-future reality, regulators are belatedly recognising that banning the technology rather than its externalities was perhaps not their finest policy moment.
Original link US considers ending ban on supersonic flights | Semafor
More analysis Sensorium link 13117
- More: FAA proposes lifting 53-year ban on supersonic overland flights
  The US FAA has proposed overturning its 1973 prohibition on commercial supersonic overland flights, replacing it with a noise-based standard requiring sonic boom overpressure below 0.53 kg/m²—a metric critics argue measures physics rather than human annoyance, and one that conveniently aligns with demonstrations by Colorado startup Boom Supersonic’s XB-1 aircraft using “Mach cutoff” techniques to refract shockwaves upward.
  Original link FAA proposal: Supersonic airliners can fly over US cities if they’re quiet - Ars Technica
  More analysis Sensorium link 13213

BCIs and Neuro Tech

🧠⚡🇨🇳Chinese chip models brain structures 478x faster than Nvidia A100
Chinese researchers from Peking University and the Chinese Academy of Sciences have built a 40-nanometre computing-in-memory chip that reconstructs the brain’s folded cortical surface in under half a second—50 to 478 times faster than Nvidia A100 GPU systems for the same task. The team’s cleverest move was treating “conductance drift” in phase-change memristors, normally considered a hardware defect, as a computational feature for neural dynamical processing, effectively turning a bug into a brain-modelling superpower.
Original link Chinese brain-mimicking chip outpaces NVIDIA GPU by up to 478x
More analysis Sensorium link 13201

Health Tech

🧬💉🛰️ Satellite Livers: MIT engineers develop injectable “mini livers” to replace transplants
MIT engineers have developed injectable “mini livers” — hepatocytes bundled with hydrogel microspheres that survived for eight weeks in mice while performing hundreds of vital liver functions, releasing specialised proteins directly into circulation. These “satellite livers” could one day replace or bridge to full organ transplants, though the current need for immunosuppressive drugs remains a hurdle the lab is actively working to eliminate with “stealthy” hepatocytes.
Original link Scientists Create Tiny “Mini Livers” That Could One Day Replace Liver Transplants
More analysis Sensorium link 13202

Bio Tech

🔬DISCO-seq maps 3D gene expression across whole intact organs
Researchers from Helmholtz Munich, the Broad Institute, ETH Zürich, and the Sanger Institute have unveiled DISCO-seq, a tissue-clearing chemistry that marries whole-organism 3D imaging with single-cell transcriptomics—yielding RNA quality on par with fresh tissue while preserving complete spatial context. Conventional methods sample less than 0.001% of an organism in 2D slices, but DISCO-seq demonstrated its chops by mapping spatially distinct glioblastoma microenvironments in mouse brains and performing unbiased whole-body tracking of SARS-CoV-2 S1 protein deposition with region-specific transcriptomic profiling.
Original link DISCO-seq: 3D single-cell transcriptomics of intact biological systems | bioRxiv
More analysis Sensorium link 13158

🧬🎯💊 LipiGo: DNA-lipid hybrid nanoparticle enables targeted drug delivery beyond the liver
Researchers at Helmholtz Munich have engineered LipiGo, a hybrid nanocarrier that embeds short single-stranded DNA molecules into lipid nanoparticles, redirecting mRNA delivery beyond the liver to lymphoid organs like the spleen while enabling modular, active cell-specific targeting via DNA hybridisation—demonstrated successfully on white adipocytes. The platform’s dual-purpose design uses DNA both as a structural biodistribution redirector and as a programmable docking system for targeting ligands, validated through whole-body tissue clearing and AI-based imaging analysis.
Original link LipiGo: A Versatile DNA-Lipid Nanoparticle Hybrid for Precision Drug Delivery | bioRxiv
More analysis Sensorium link 13159

Climate Tech

☀️⚡CIGS-perovskite tandem solar cell sets 25.5% efficiency record
A Berlin-based research team has certified a new 25.5% efficiency record for CIGS-perovskite tandem solar cells, edging past the previous 24.6% benchmark and earning a spot in the prestigious “Green Tables” published in *Joule*. The team optimised everything from band gaps to Buckminsterfullerene deposition rates—and claims in-house testing is already hitting 27.5%, making this record less a finish line and more a particularly well-documented waypoint.
Original link Tandem solar cell sets 25.5% efficiency record with CIGS-perovskite design
More analysis Sensorium link 13078

Materials Science

🧱🌍📉 Rebond 300: The Carbon-Negative Concrete That Turns Buildings Into Stone-Age Carbon Vaults
Dutch materials company Paebbl has launched Rebond 300, the world’s first carbon-negative construction material—a supplementary cementitious powder that replaces up to 30% of Portland cement while locking away -149 kg of CO2 per tonne through accelerated mineralization that mimics nature’s geological CO2-to-stone process on a factory timeline. Unlike the muddy-grey slag substitutes that architects have historically banished to basement foundations, Rebond 300’s near-white finish means it can proudly bare its face on facades—and it slots into existing ready-mix workflows without a single new piece of equipment.
Original link World’s first carbon-negative building material traps CO2 in walls
More analysis Sensorium link 13111

Deep science

⚛️🔬✨ HL-LHC: CERN begins $1.5B upgrade to unlock Higgs secrets
CERN’s Large Hadron Collider signed off on June 27, 2026, kicking off a US$1.5 billion, four-year transformation into the High-Luminosity LHC — same 27-kilometre tunnel, same 13.6 TeV beam energy, but with world-first niobium-tin focusing magnets, superconducting crab cavities that sideways-shuffle proton bunches into head-on collisions, and machine learning triggers burned directly onto FPGAs to filter 99.9% of five billion particle interactions per second in real time. The payoff by 2030: ten times the collision rate, 380 million Higgs bosons, and humanity’s first crack at observing di-Higgs production — the only known window into the Higgs self-coupling constant that determines whether our universe sits in a permanently stable vacuum… or a metastable false minimum that could, in principle, tunnel to a lower-energy state at some cosmologically inconvenient future date.
Original link Large Hadron Collider Powers Down: $1.5B Upgrade to Unlock Higgs Secrets
More analysis Sensorium link 13051

🥀🌍⏳ The Last Salad: Earth’s plants have 1.9 billion years left, study predicts
New modelling from Blue Marble Space researchers suggests Earth’s plant life could persist for nearly 1.9 billion years — longer than some previous estimates — before either rising solar radiation cooks the biosphere or plummeting CO₂ levels starve it. The two trajectories converge on a grim deadline somewhere around 1.84–1.87 billion years CE, after which even the hardiest photosynthesisers will have had their last gasp, with oceans vaporising roughly 300 million years prior.
Original link Earth could stay green for nearly 1.9 billion years
More analysis Sensorium link 13107

🏭AI and Tech industry news

Just links in this section… they get enough coverage, already…

🫧Bubble Chronicles

Local models signal the beginning of the end for Big AI
Original link The Beginning of the End for Big AI - by Nicolas Colin
More analysis Sensorium link 13055

Retirement funds quietly financing the AI data center boom
Original link AI Data Centers: The Real Reason They’re Going Up Everywhere
More analysis Sensorium link 13172

OpenAI and Anthropic face major hurdles going public
Original link Why OpenAI and Anthropic may struggle to float
More analysis Sensorium link 13189

US Treasury secretly warns AI bubble could mirror dotcom crash
Original link Treasury Has an Internal Report Warning About the Dangers of an AI Bubble - NOTUS — News of the United States
More analysis Sensorium link 13250

Nvidia

NVIDIA slashes DeepSeek v4 token costs 5x with Blackwell
Original link NVIDIA Slashes DeepSeek v4 Token Costs By Up To 5x Just One Month After Launch, Through Pure Blackwell Software Tuning
More analysis Sensorium link 13079

Nvidia AI server delay rattles Asian tech stocks
Original link Nvidia AI Server Delay Report Sends Asian Tech Stocks Sliding - Bloomberg
More analysis Sensorium link 13174

Nvidia launches revenue-sharing model to double-dip on AI cloud profits
Original link Nvidia offers to take a cut of AI cloud revenue on top of hardware sales in new optional financing vehicle — trades tokens for revenue cut | Tom’s Hardware
More analysis Sensorium link 13182

Palantir and Nvidia launch air-gapped AI stack to end token billing
Original link Palantir and Nvidia Launch Air-Gapped AI Stack as Token Billing Cracks Enterprise Budgets
More analysis Sensorium link 13149

OpenAI

💰🤖🇺🇸 OpenAI Floats 5% Equity Gift to Uncle Sam, Sparking Debate Over Who Really Benefits
Seriously…
Coverage:
More analysis Sensorium story 373
Sam Altman pushes for global AI safety standards forum
Original link Sam Altman calls for US-led international forum to set global AI standards - SiliconANGLE
More analysis Sensorium link 13212

ChatGPT usage surges globally, led by Africa and Asia
Original link How ChatGPT adoption has expanded | OpenAI
More analysis Sensorium link 13215

OpenAI slashes inference costs 50% using software alone
Original link OpenAI Halves Inference Costs With Software Alone: GPUs Drop to Hundreds
More analysis Sensorium link 13227

Anthropic

🔓🤖🏛️ Trump drops restrictions on Anthropic’s Mythos and Fable models
Coverage:
More analysis Sensorium story 370
Claude Code’s 1,000-agent Dynamic Workflows now reach all paid plans
Original link Claude Code Dynamic Workflows Go GA: Pro Users Can Now Spawn 1,000 Parallel Agents
More analysis Sensorium link 13086

Anthropic adds AI safeguards to lift Fable 5 export ban
Original link Anthropic Added a New Security Measure to Get Back Into the Trump Administration’s Good Graces | WIRED
More analysis Sensorium link 13099

Claude Fable 5’s safety classifier silently downgrades 75% of coding tasks
Original link Claude Fable 5 Debugging Scores Drop 70%: Safety Classifier Reroutes Tasks to Weaker Fallback Model
More analysis Sensorium link 13146

Anthropic cracks down on Chinese firms bypassing Claude restrictions
Original link Anthropic moves to close loopholes that allow Chinese access to Claude
More analysis Sensorium link 13147

Alibaba bans Claude Code over Anthropic’s hidden user tracking
Original link Alibaba bans Claude Code over hidden Chinese user tracking
More analysis Sensorium link 13138

Emails reveal Pentagon blacklisted Anthropic over autonomous weapons limits
Original link Pentagon Blacklisted Anthropic Over Autonomous Weapons Limits: Emails Reveal “Very Close” Talks
More analysis Sensorium link 13143

Anthropic eyes Samsung 2nm chip to cut Claude inference costs
Original link Anthropic in Talks With Samsung to Build Custom AI Chip, Aiming at 2nm Process
More analysis Sensorium link 13150

Anthropic plans to develop its own drugs using AI
Original link Anthropic wants to develop its own drugs | The Verge
More analysis Sensorium link 13195

California government gets Anthropic’s Claude AI at 50% discount
Original link Newsom strikes Anthropic deal to get California government half price Claude AI access | TechRadar
More analysis Sensorium link 13247

Google

Google loses appeal against record €4.1bn EU Android fine
Original link Google loses appeal against record €4.1bn EU fine over Android
More analysis Sensorium link 13124

Microsoft

Azure Linux 4.0 arrives to challenge Windows Server
Original link Microsoft’s new Azure Linux 4.0 is here, and it could replace Windows Server in the enterprise | ZDNET
More analysis Sensorium link 13102

🏢🤖💰 Microsoft Bets US$2.5 Billion on Forward-Deployed AI Services with Frontier Company Launch
Coverage:
- Microsoft launches AI-focused professional services business with $2.5B investment - SiliconANGLE
- Microsoft launches its own AI deployment company with $2.5 billion commitment | TechCrunch
More analysis Sensorium story 377

Apple

Apple and EU hold constructive talks over Siri AI
Original link Apple’s Tim Cook holds ‘constructive’ talks with EU tech chief over ‘Siri AI’
More analysis Sensorium link 13120

India probes Tata breach exposing iPhone 18 Pro’s secret dual-modem strategy
Original link iPhone 18 Pro Leak: India Opens Criminal Probe as Stolen Files Reveal C2 Modem Plans
More analysis Sensorium link 13211

AliBaba

Alibaba bans employees from using Anthropic’s Claude Code
Original link Alibaba reportedly bans employees from using Claude Code | TechCrunch
More analysis Sensorium link 13142

Tencent

Tencent’s Hy3 goes Apache 2.0, beats larger rivals on agent tasks
Original link Tencent’s Apache-licensed Hy3 takes on GLM-5.2 at half the size — and wins everywhere except coding | VentureBeat
More analysis Sensorium link 13184

SpaceX

SpaceX slashes Starlink prices amid xAI data center backlash
Original link SpaceX offers half-price Starlink internet plans in Memphis to quell backlash to its Colossus data centers — as it promises to ‘continue to invest in the area’ | TechRadar
More analysis Sensorium link 13059

Elon Musk denies report of SpaceX AI phone prototype
Original link Elon Musk denies a report about SpaceX’s AI phone prototype | The Verge
More analysis Sensorium link 13104

SpaceX reportedly tests handheld AI device, Musk denies it
Original link Report Suggests SpaceX Is Testing A Handheld AI Device, Musk Says It’s ‘Utterly False’
More analysis Sensorium link 13068

Startups and Investment Deals

How Zurich became the world’s secret AI R&D hub
Original link Building tech in the world’s secret R&D hub | MIT Technology Review
More analysis Sensorium link 13053

Tycoon bets $30M on AI alternative to Microsoft Office
Original link Indian tech tycoon bets $30M of his own money to build AI alternative to Microsoft Office | TechCrunch
More analysis Sensorium link 13125

Palantir CEO blasts AI industry’s “wealth tax” on businesses
Original link Palantir CEO Alex Karp doesn’t hold back in interview as he rails against AI industry - SiliconANGLE
More analysis Sensorium link 13224

Even Realities raises $150M for AI-powered smart glasses
Original link Smart glasses startup Even Realities raises $150M - SiliconANGLE
More analysis Sensorium link 13200

Venice AI hits unicorn status with $65M privacy-focused raise
Original link Venice AI becomes a unicorn with $65M Series A as its privacy-first AI platform takes off | TechCrunch
More analysis Sensorium link 13239

⏳ Zeitgeist

Just a few links from outside tech bubble…

Climate

In case you momentarily forgot that Climate Change is happening:
(Source:)
Simon Kuestenmacher@simongerman600
This animated version of the NASA climate spiral is worth watching until the end when the graphic turns 3D. Source: reddit.com/r/Damnthatsint…
9:00 PM · Jul 1, 2026 · 52.8K Views
27 Replies · 259 Reposts · 612 Likes
🌊🔥📈 Boiling Point: Record ocean heat and El Niño signal more extreme conditions ahead
Global ocean temperatures have shattered June records, with the average sea surface temperature hitting nearly 21°C—a staggering thermal load equivalent to detonating 12 Hiroshima-scale bombs every second throughout 2025. As a formidable El Niño forms in the Pacific, we’re staring down a future of supercharged cyclones and marine heatwaves, all while the US administration plays Russian roulette with vital climate monitoring networks.
Original link The world’s oceans are the hottest on record for June—and El Niño is set to turn up the heat even more
More analysis Sensorium link 13169

🥵🏚️🇫🇷 Survival Blankets as Window Blinds: France’s record heat wave kills thousands, deaths up 30%
France recorded a 29.1% spike in mortality—2,025 excess deaths—during a record-breaking June heat wave that pushed temperatures above 40°C across more than 40% of the country, with the Paris region suffering a staggering 62% increase. With one in three French homes poorly insulated against heat, residents resorted to taping survival blankets over windows, while the government of PM Sébastien Lecornu faces a no-confidence vote over alleged unpreparedness—a political heat of its own kind.
Original link France deaths rose by 30% during heat wave
More analysis Sensorium link 13193

Health

🚢🦠✅ Operation Hondius: WHO declares hantavirus cruise ship outbreak officially over
The WHO has declared the hantavirus outbreak aboard the Dutch-flagged cruise ship MV Hondius officially over, with all 13 cases resolved and 650+ contacts across 33 countries successfully traced through quarantine. The culprit was the Andes virus — the only known hantavirus capable of human-to-human transmission — making this less a routine rodent-borne infection and more a stress test for global pandemic infrastructure.
Original link The hantavirus outbreak is over, with a toll of 13 cases and three deaths | Live Science
More analysis Sensorium link 13199

🌊💩Flooded zone

Just had to note this:

Trump’s memecoin buyers lost $3.8B while he profited
Original link Trump Memecoin Puts 1M Holders at Losses of $3.8B
More analysis Sensorium link 13243

🧠Mind expanding

🧮AI pushes markets from competition toward algorithmic collusion
A new paper from Philip Maymin proves that market competition persists not because of antitrust law or regulatory oversight, but because firms lack the computational power to sustain collusion — specifically, if P≠NP, the collusion detection problem is computationally infeasible, making punishment threats non-credible and collusion unstable. Combined with Maymin’s 2011 result that market efficiency requires P=NP, this yields a fundamental impossibility: markets can be informationally efficient or competitive, but not both — and AI, by expanding firms’ effective computational capabilities, is actively pushing markets from the competitive regime toward the collusive one, explaining the empirical emergence of algorithmic collusion without explicit coordination.
Original link Markets are competitive if and only if P != NP
More analysis Sensorium link 13157
🏛️💻⚔️ Silicon Empires
Nick Srnicek’s new book traces how a handful of Big Tech firms are consolidating power across the entire AI supply chain, reshaping global economics and politics in their image.
Original link How a handful of tech firms are reshaping the global economy | Semafor
More analysis Sensorium link 13210

💭Meme stream

🏔️🌿🏘️ Nieuw Bergen
MVRDV has delivered 237 homes in Eindhoven’s historic Bergen district through a cleverly staggered development that rises like a miniature mountain range, with dramatically angled green roofs that pull daylight deep into the apartments while integrating solar panels, communal terraces, and a greenhouse. The project fuses adaptive reuse of two former police stations with five new buildings — including 54 social housing units — proving that density and neighbourhood character aren’t mutually exclusive.
Original link Green roofs reshape Dutch housing in historic Eindhoven
More analysis Sensorium link 13151

✈️📄🏆 ICARUS: Students in Pisa break world record with 20-metre paper plane
A team of students from the University of Pisa has set a new Guinness World Record for the largest paper plane, a 7m-long, 20.04m-wingspan glider weighing just 28.49kg that flew 59m — built using genuine aeronautical engineering principles including spars, ribs, and a honeycomb paper structure that would make any aerospace engineer nod approvingly. The project consumed roughly 300kg of paper and 60kg of glue, yet the real marvel is that a single person launched it from a 3m platform, proving that sometimes the most absurd engineering challenges yield the most elegant solutions.
Original link Students set Guinness World Record with giant paper plane
More analysis Sensorium link 13234
From He-Man to Ulysses 31
Showing my age here… who remembers these? Some real blasts from the very distant past here…
- He-Man and beyond: 20 sci-fi cartoons (some iconic, some weird) that transported ‘80s and ‘90s kids to strange new worlds
Modern work life

An update on our slide into feudalism
Is Substack the place?
Let’s find out, shall we?

🙏🙏🙏 Thanks as always to everyone who takes the time to get in touch with links and feedback.

Reminder to paid subscribers, email me to get an invite code to the Sensorium Beta:

View more detailed analysis of all the links referenced in this newsletter

Download all newsletters into your own knowledge graph

Access the entire searchable Memia archive

AND check out next week’s newsletter in real time as it’s being compiled

Ngā mihi

Ben

🧛AI vampires

Discussion about this post

Ready for more?

Memia #2026.27: AI vampires🧛 UWORLD U1🤖💙 compression for human🗜️🧠 j-lens🪞 orbital data embassy🛰️💾 AI allegory steganography📖 summer of ludd📵 tinyretropad📝 balcony solar 🌅🔌 stormwall☀️🛡️

Found my tribe🤖 🤖

🗞️Weekly roundup

Found my tribe🤖 🤖

🥄💀📈Bending Spoons, extracting rent

🧛AI vampires

🌟Picks of the week

📈The week in AI and Tech

AI safety

Governance and Policy

Regulation

Security

Sovereignty and Geopolitics

Society

The Economy

Business

Environment

Entertainment

🆕 AI releases

🥼 AI research

🔮[Weak] signals

Consumer Tech

Chips and Computer Hardware

Cybersecurity

(Tele)communications

Robotics

Autonomy and Drones

Military Tech

Space

Crypto

Energy

Transport

BCIs and Neuro Tech

Health Tech

Bio Tech

Climate Tech

Materials Science

Deep science

🏭AI and Tech industry news

🫧Bubble Chronicles

Nvidia

OpenAI

Anthropic

Google

Microsoft

Apple

Amazon

Meta

xAI

ByteDance (TikTok)

Samsung

Huawei

AliBaba

Tencent

SpaceX

Startups and Investment Deals

⏳ Zeitgeist

Climate

Health

🌊💩Flooded zone

🧠Mind expanding

💭Meme stream

Discussion about this post

Ready for more?