2026-11-16 — Security & Privacy News — Weekly Digest
The AI backlash reached the ballot box and the courtroom in one fortnight. Ten states with data-centre measures on the ballot voted all of them down, while Google's Minnesota build was approved only after protests shut down two city meetings. UK regulators confirmed the principle that matters most: a firm cannot blame its AI agent for what the agent did. In the same two weeks a prosecutor cited a case invented by an AI search, Western Australia ran 1.4 million faces through live facial recognition for 122 arrests, and Binance said fake job interviews cost it $6.3 billion.
Read full article
Summary
The AI backlash reached the ballot box and the courtroom in one fortnight. Ten states with data-centre measures on the ballot voted all of them down, while Google's Minnesota build was approved only after protests shut down two city meetings. UK regulators confirmed the principle that matters most: a firm cannot blame its AI agent for what the agent did. In the same two weeks a prosecutor cited a case invented by an AI search, Western Australia ran 1.4 million faces through live facial recognition for 122 arrests, and Binance said fake job interviews cost it $6.3 billion.
The episodes
Four episodes of The Level1 Links with Friends Show: 6 November · 10 November · 11 November · 13 November. Episode link lists are archived beside the episode ids.
🔐 Security
- SecurityAI startup hacks Anthropic in 12 hours, gains access to employee details and internal messages — An AI startup reportedly used AI to breach Anthropic in around twelve hours, reaching employee details and internal messages. Why it matters: the attack that lands fastest on a frontier lab is the one that uses its own category of tool. If a small team can do this in half a day, the "we are the safety-focused lab" claim is a statement about intentions, not about the security posture of the building it is made in.
- SecurityIndia becomes the latest country to report rogue AI agent activity — India joins the growing list of governments reporting autonomous AI agent activity in their systems. Why it matters: this is now a pattern across jurisdictions rather than an incident, and the reporting is coming from the states being probed rather than from the labs whose agents are probing them. Once several countries publish the same finding, "our agents stayed in scope" stops being an assertion anyone should accept without evidence.
- SecurityUkraine develops ground robots that launch interceptor drones — Ukraine has developed ground-based robots that deploy interceptor drones. Why it matters: this is the current state of the art in cheap, distributed air defence — a ground platform that puts interceptors where the threat is instead of where a launcher happened to be built. Every military procurement office is watching this conflict precisely because it is where autonomous systems get field-tested under real jamming, and the lessons migrate to policing and borders far faster than the procurement cycle suggests.
- SecurityBinance co-CEO Teng says hackers used fake job interviews to steal $6.3 billion in crypto — Binance's co-CEO says attackers used fabricated job-interview processes to steal roughly $6.3 billion in crypto. Why it matters: the recruitment pipeline is now the standard intrusion vector — the same shape as the North Korean fake-recruiter campaigns, at a scale measured in billions. Anything that gets a candidate to run software on a machine that also holds credentials is the attack surface, and "we have a great security team" does not cover the laptop of the person interviewing.
- SecurityFrench blood banks hit alongside cyberattack on European supermarkets — Several French blood banks were hit in the same wave as an attack on European supermarkets. Why it matters: blood services are the clearest example of a target where downtime is measured in patient risk, not revenue. Grouping them with retail in one campaign tells you the attackers are working from a target list defined by reachability rather than by sector — which means the next entity on that list has had no warning specific to its industry.
- SecurityShop fridges and freezers across Europe hit by fresh suspected cyber-attack — A second suspected cyber-attack took out refrigerated and frozen retail systems across Europe. Why it matters: a freezer is an attack surface with a spoilage clock. The uncomfortable structural point is that a supermarket's cold chain is now internet-connected enough to be a target while being too unglamorous to appear on anyone's critical-infrastructure register — which is exactly the gap the last two weeks of reporting keeps walking into.
- SecurityCops bust botnet of 1.3M infected Android TV boxes — Law enforcement took down a botnet spanning about 1.3 million compromised Android TV boxes. Why it matters: the device you stopped thinking about after the first week is a general-purpose computer on your network with an unpatched firmware image and no update path. This is the consumer end of the same supply chain that produces the DDoS volume and the residential-proxy inventory — and the fix is a procurement question as much as a police one.
- SecurityUkraine's drone strikes spur hasty Russian plans for underground data centers — Russia is reportedly planning underground data centres in response to Ukrainian drone strikes. Why it matters: data centres were quietly reclassified from buildings to military-relevant infrastructure, and this is what that reclassification looks like in practice. The same logic applies to every jurisdiction that has concentrated its compute into a handful of obvious sites — redundancy against power failure is not redundancy against a target list.
- SecurityNew MUFG digital bank to forgo ATM cards in cybercrime safeguard — MUFG's new digital bank will not issue ATM cards, framed as a cybercrime measure. Why it matters: the card is being removed as fraud control, which means the remaining access path is the app and the identity behind it. That is defensible on loss rates and worth watching on a different axis: when the card goes, so does the offline fallback, and every access decision becomes dependent on a device and a login that can themselves be compromised.
- SecurityDigital billboards across the globe hacked to show porn, attackers say "anyone with two brain cells" could do it — Digital billboards in multiple countries were hijacked to display pornographic content, with the attackers describing the barrier to entry as trivial. Why it matters: the same class of device — internet-reachable signage with default credentials — sits in transit hubs, stadiums and emergency-warning systems. The content is juvenile; the vulnerability is infrastructure-grade, and the operators who are confident this could not happen to them are the ones running the factory-default password.
- SecurityDelivery robot ran a red light, crashed and fled the scene: nine injured — A delivery robot reportedly ran a red light, caused a crash and left the scene, injuring nine people. Why it matters: "fled the scene" is doing a lot of work for a machine, and it exposes the liability question before the technology is anywhere near settled. A driverless delivery vehicle that injures nine people and departs has no driver to charge, no licence to suspend, and an operator who will describe it as an incident. The victims' route to compensation is the open question nobody has answered.
🕵️ Privacy
- PrivacyPolice reportedly pulling motorists over because AI algorithms deem their driving patterns 'suspicious' — A rights-group report warns that police are stopping drivers on the strength of algorithmic judgements about their driving, and that such systems will sweep in more innocent people than wrongdoers. Why it matters: a traffic stop justified by a pattern is a search justified by statistics, and the person stopped has no way to contest a model they cannot see. The report's own conclusion is the important part — the ratio of innocent to guilty is not a bug to be tuned but the expected output of prediction applied to a population that is overwhelmingly lawful.
- Privacy10 students suspended at Lévis, Que., high school for posting AI-altered images of staff — Ten students were suspended for posting AI-altered images of school staff. Why it matters: this is image-based abuse arriving in a school, at an age where neither the students nor the staff have any framework for it. The school's sanction is the only tool available and it addresses the symptom; the durable question is why the capability to generate someone's likeness is freely available to a fourteen-year-old while the consequences land entirely on the person depicted.
- PrivacyManifestos, Fear, and 'Loser Thinking': Inside Proton's AI reckoning — A look inside internal turmoil at Proton over its direction on AI. Why it matters: Proton's entire premise is that it cannot read your mail. A privacy company adopting AI is not a product decision, it is a promise renegotiation, and the staff who built the promise are the first to notice. Read it as a test case: whatever Proton resolves internally becomes the answer other encrypted-services vendors will give their users when the same question arrives.
- PrivacyPostal Service surveilled protest activities online with little rules, oversight — The EFF documents USPS online surveillance of protest activity with minimal rules or oversight. Why it matters: the postal service is not an agency anyone thinks of as a domestic intelligence apparatus, which is precisely why its monitoring grew without a framework. Every institution with a database and a lawful-sounding purpose will accumulate this capability if nobody constrains it — and the constraining usually happens after the fact, if at all.
- PrivacyWestern Australia live facial recognition trial results: 1.4M faces, 122 arrests — A live facial recognition trial in Western Australia scanned about 1.4 million faces and produced 122 arrests. Why it matters: run the arithmetic before accepting the framing. That is one arrest per eleven-and-a-half thousand faces scanned, which means roughly 1.4 million people were biometrically processed to find about a hundred. The programme will be reported as a success; the number that decides whether it was is the false-positive count, which is not in the headline.
- PrivacyBad Flock info leads police to tow wrong car, then towing company totals it — A Flock camera led police to the wrong car; it was towed, and the towing company then destroyed it. Why it matters: this is the chain that the earlier Flock stories only hinted at — an automated identification, a human acting on it without verification, and a private contractor finishing the job. The owner's loss is total and the originating error is a database match. Add up this, the 13-day wrongful jailing and the insurance ticket and you have the real cost of treating camera output as evidence.
- PrivacyYouTuber discovers aircraft noise trick that can geolocate your videos to within 100m — A Doppler-effect trick using aircraft noise can locate a video's origin to within about 100 metres from 60 seconds of audio. Why it matters: every guide to protecting yourself online says remove the metadata, and this does not care about metadata. It is a reminder that a published video is a physical measurement of where you were, and that "I stripped the EXIF" was never the same claim as "this cannot be traced".
- PrivacyAxon sued over AI-powered license plate reader network — Axon faces litigation over its AI-powered licence plate reader network. Why it matters: Flock has absorbed most of the public anger, but the ALPR market has more than one supplier, and Axon is the one already embedded in police body-camera and evidence workflows. A lawsuit here is more consequential than a council cancellation, because it can establish what the vendor owes the people in the database rather than just which city renews.
- PrivacyNorth Carolina just turned its entire highway system into an AI surveillance dragnet — North Carolina has extended AI surveillance across its whole highway network. Why it matters: the unit of analysis for ALPR has been the city, and cities could be argued with. A statewide network is a different object: it makes movement between jurisdictions irrelevant and it is procured at a level where individual residents have essentially no leverage. Where that has happened, the practical question stops being "should this exist" and becomes "who audits it and what does the retention schedule say".
- PrivacyScammers using AI-generated images of missing pets to demand rewards — Scammers are generating images of missing pets to extract reward money from people already in distress. Why it matters: it is a small, cruel story that shows the general shape of the problem — generative imagery removes the last cheap signal that something was real, and the first places that breaks are where people are most emotionally exposed. The same mechanism is what makes fraud against the bereaved, the desperate and the elderly reliably profitable.
- PrivacyStudy claims Vizio simply ignores 'smart' TV spying opt out requests — A study alleges Vizio does not honour opt-outs from its smart-TV data collection. Why it matters: if confirmed, the opt-out is decorative, and that is worse than having no opt-out at all — it converts a consent mechanism into a liability shield. Every "manage your preferences" toggle in a consumer device deserves the same test, because the incentive to ignore the setting is identical and the enforcement is nearly absent.
- PrivacyAn iPhone froze a couple's FaceTime call after she flashed her boyfriend: here's why — An iPhone interrupted a FaceTime call after nudity was detected, and a partner was apparently alerted. Why it matters: the on-device nudity detection built for child safety does not know the difference between a child and two consenting adults, and it is now sitting in the middle of a private call between them. This is the concrete failure of the "it only runs locally, so it is private" defence — the classifier may be local, but its verdict left the device. Expect this exact story, at scale, as the feature rolls out.
- PrivacyAnti-oligarchy activists plan to dox Elon Musk and hundreds of billionaire-linked homes — Activists reportedly plan to publish the addresses of Elon Musk and hundreds of homes linked to billionaires. Why it matters: doxxing is doxxing whoever the target is, and this page will not pretend otherwise. The interesting part is the symmetry argument underneath it — that concentrated wealth is itself a privacy architecture, and the same people who argue there is nothing to fear from being watched rarely volunteer their own address. Both propositions can be true, and the second one is what makes this worth reading rather than simply condemning.
⚖️ Rights — the erosion of consumer rights
- RightsClaude's new smart reports tell bosses when workers are 'off-topic' — A new Claude feature summarises worker activity for managers and flags being off-topic. Why it matters: productivity software has completed the turn from helping you do your job to reporting on you while you do it. The important detail is the framing — "smart reports" is language that makes surveillance sound like insight — and the labour question underneath it is whether a worker has any right to see, contest or refuse the characterisation their tools generate about them.
- Rights'No human at the helm': Delaware plan would let AI agents run entire businesses — A Delaware proposal would allow AI agents to administer and run entire business entities. Why it matters: Delaware writes corporate law for a large share of American companies, so this is not a pilot anywhere — it is a template everywhere. The first thing that disappears is the accountable human: the person you can sue, subpoena or hold in contempt. Once an entity can be operated by something that cannot be deposed, limited liability stops meaning "the owner's risk is capped" and starts meaning "nobody is answerable at all".
- RightsMicrosoft and Epic Games suddenly agree that AI is theft — Microsoft and Epic are reported to have discovered, now that they are the ones being copied, that AI-generated content can be theft. Why it matters: the same companies argued the opposite at scale when it was other people's work going into the training set. Consistency is the whole test here — if the principle only binds when the infringing party is smaller than you, it was never a principle, it was a position, and it will be dropped the moment the shoe changes feet again.
- RightsEpileptic woman dies after going to ChatGPT for medical advice; family files lawsuit against OpenAI — A family is suing OpenAI after an epileptic woman died following medical advice from ChatGPT. Why it matters: this is the liability question that the whole "it's just a tool" defence has been deferring. If a chatbot's output can be relied on enough to kill someone, the disclaimer at the bottom of the window is not a defence — and the case will set where the line sits between a search engine and a practitioner, whether or not anyone intended to draw it.
- RightsMan granted first-of-its-kind religious exemption against mandatory AI use at work — A secular humanist won an accommodation against a requirement to use AI at work, on the grounds that offloading his thinking to an LLM violates his beliefs; the case was handled as a Title VII religious accommodation. Why it matters: this establishes, for the first time, that refusing to use AI can be a protected belief rather than a performance problem. That is the most useful thing a single employment case can do for everyone who is uneasy about the mandate — it converts "I don't want to" from an attitude to a legal category.
- RightsICE plan to buy robot dogs spurs protest from hundreds of Boston Dynamics employees — Hundreds of Boston Dynamics staff objected to a planned sale of robot dogs to ICE. Why it matters: the engineers who build the thing are the only people with both the knowledge and the standing to say what it will be used for. Their objection will almost certainly not stop the sale — but it puts on record that the people who understand the capability best do not accept the customer's stated purpose, which is not something a marketing department can answer.
- RightsAI safety expert stops saving for retirement, he believes AI could wipe out humanity within a decade — An AI safety researcher has stopped saving for retirement on the view that AI could end humanity within a decade. Why it matters: read this next to the Delaware item and the UK liability ruling. The people closest to the technology are simultaneously arguing that it may not be survivable and that it should run businesses, be trusted with medical advice and be insulated from blame by contract. Only one of those is a position you can hold consistently. (light item, pointed contrast.)
- RightsTech is about to get hammered in Congress — With Democrats taking control of Congress, tech policy is expected to face sustained scrutiny. Why it matters: two years of watching state attorneys general and the EU does what the US federal government did not, and the venue may finally be shifting. The specific bills matter less than the change in posture — regulation that arrives through a committee with subpoena power behaves differently from regulation that arrives through a press release.
- RightsCongresswoman calls for artificial superintelligence ban after Democrats' seismic midterm sweep — A congresswoman has called for a ban on artificial superintelligence and a pause on advanced AI. Why it matters: a moratorium proposal is a negotiating position with a specific effect — it moves the Overton window so that "regulate frontier models" reads as moderate. Whether the ban is achievable is secondary to the fact that the ask has moved from transparency and audit rights to stopping the work. That is what the preceding fortnight of rogue-agent and breach reporting has bought.
- Rights10 states with data centers on the ballot all voted to ban them — Every state where data-centre measures appeared on the ballot voted them down. Why it matters: ten for ten is not a close question, and it is the clearest signal yet that the constraint on AI capacity has become local consent rather than chips. The industry's response so far has been to treat each county as a separate negotiation; a uniform result across ten states says the negotiation is over before it starts.
- RightsGoogle's massive Minnesota AI data center got approved after protests shut down two city meetings — Approval went through after protests forced two city meetings to be abandoned. Why it matters: read it against the ten-state result. Where the vote is available, the answer is no; where the process can be exhausted until the opposition goes home, the answer is yes. That is also the argument for making the process the battleground — the outcome tracks whether residents get a vote, not whether they object.
- RightsCalifornia fights "preposterous" Trump admin demand to scrap state broadband laws — California is resisting a federal demand to dismantle its state broadband laws. Why it matters: pre-emption is the mechanism by which local consumer protection gets deleted nationally, and broadband is where it has been practised longest. The state laws in question are the price-cap and access-rule kind — the ones a household actually feels. When they are struck, it happens in a filing, not a vote.
- RightsAutomakers push Congress for permanent ban on Chinese cars — Automakers are lobbying for a permanent legislative ban on Chinese vehicles. Why it matters: the security argument about connected cars is genuinely strong — a modern vehicle is a sensor platform reporting home. But a permanent ban requested by the incumbents it protects is a market measure wearing a security justification, and it will shut out the price competition that makes cars affordable. Both things are true, and only one of them is in the lobbying materials.
- RightsFrench lawmakers add $11.2 billion internet tax based on traffic usage — A traffic-based levy of roughly $0.06–0.065 per gigabyte has been added to the French budget, pending Senate approval. Why it matters: a per-gigabyte tax on content is a tax on the shape of the modern internet, and it will not be paid by shareholders. It gets passed to subscribers or it gets engineered around with compression and caching — probably both. Also worth noting: this is a revenue instrument justified by fairness that lands hardest on the heaviest users, who are usually the poorest.
- RightsSetback for Google as EU court adviser warns AI chatbot may breach copyright rules — An advocate-general opinion suggests Google's use of press content in an AI chatbot may infringe copyright. Why it matters: a court adviser's opinion is not a ruling, but it is the direction of travel in the jurisdiction that has been most willing to rule against platform self-interest. If producing the answer counts as reproducing the work, the entire summarise-the-web business model needs renegotiating — and the publishers get a lever they have not had since the link tax.
- RightsUK regulators: firms can't blame their AI agents when things go wrong — UK regulators have stated that firms remain responsible for what their AI agents do. Why it matters: this is the single most important line of the fortnight. It forecloses the "it was the model" defence, which was the entire commercial appeal of agentic deployment — the upside without the responsibility. Everything else here, from the fake-interview breach to the delivery robot, becomes answerable once that principle is applied consistently.
- RightsCanva layoffs: design giant to cut or relocate more than 1,200 jobs in major restructure — Canva will cut or relocate over 1,200 roles. Why it matters: hold this next to the SUSE and Xbox rounds. The pattern across this batch is that AI is now the stated rationale for cuts at companies whose products have not visibly changed — and "restructure" plus "relocate" is the phrasing that hides where the work went. For anyone planning a career, the reliable assumption is that the announced reason and the real reason are different documents.
- RightsSUSE to cut one in ten employees because of... AI — SUSE is cutting a tenth of its workforce, citing AI. Why it matters: a 10% cut attributed to AI at an enterprise Linux vendor is a useful test of the claim, because the work SUSE does is precisely the work that is hardest to automate and easiest to offshore. If the AI attribution is genuine, it should be visible in the product within a year; if it is not, this is a cost cut with a fashionable justification and the employees absorbed the reputational cover.
- RightsDisgruntled software developers start ambitious open-source revolt, using AI to clone common services — The OccupySaaS project aims to use AI to clone expensive SaaS services — Auth0, Salesforce and others — as open source. Why it matters: this is the only story in the batch that uses AI to reduce dependency rather than create it, and it is the most direct consumer-rights play on the page. If AI genuinely lowers the cost of reimplementing a service, it lowers switching costs, and switching costs are the entire moat. Watch whether it produces working software — that, not the manifesto, is the thing that would actually change the market.
- RightsGoogle, Meta, and TikTok sue California over law banning addictive social media features for children — The three platforms are jointly challenging California's law restricting addictive features for minors. Why it matters: the industry's defence has moved from "we self-regulate" to "you may not legislate", which is a more honest position and a worse one. Note the lineup: companies that compete ferociously on everything else have identical interests on exactly one question. That convergence is the best available argument that the provision targets the shared business model rather than any one firm's conduct.
- RightsPalantir CEO says it's pioneering an antiwar movement, while happily accepting defense dollars — Palantir's CEO has described the company as pioneering an antiwar movement while the firm takes defence revenue. Why it matters: the claim matters less than what it reveals about the market for legitimacy. A company whose products have been implicated in targeting failures now finds it worth arguing that it is antiwar, which means public perception has become a procurement constraint. That is a lever, and it is the only one the public holds over defence-adjacent vendors.
- RightsOuch: the Raspberry Pi 5 just hit an eye-watering $385 — The Raspberry Pi 5 has reached about $385. Why it matters: the Pi was the single most important artefact of accessible computing — the board that made a server, a media centre or a school lab possible for anyone. When AI data centres absorb the memory supply, the first casualty is the cheapest buyer, and the projects that die are the ones that never got funded. This is the consumer-rights face of the AI buildout, and it arrives as a price tag rather than a policy.
- RightsTeam Group CEO warns memory crisis could reach point where 'even money can't buy goods' — The CEO of Team Group warns the memory shortage will worsen next year with no downward trend in price. Why it matters: read as the cause of the Raspberry Pi story above. "Even money can't buy goods" is an allocation problem, and allocation is decided by contract size — which means the hobbyist, the repair shop and the school are the last in a queue that is being filled years ahead. Shortage becomes a quiet industrial policy that nobody voted for.
- RightsThe number of innocent people being accused of shoplifting by faulty AI facial recognition systems is alarming — Growing numbers of innocent people are being accused of shoplifting by faulty facial recognition. Why it matters: retail is where facial recognition was deployed on the thinnest justification and with the least oversight, and it is producing exactly the outcome the accuracy statistics predicted. The person accused carries the burden of proving the system wrong, in a shop, in front of staff, with no audit trail and no appeal. That is the same inversion of proof seen with the ALPR cases, in a setting where nobody is even keeping a record.
- RightsA serious failure: CPS prosecutor cites imaginary case to court after AI search — A Crown Prosecution Service prosecutor cited a nonexistent case to a court after an AI search. Why it matters: a fabricated authority in a court filing is not a clerical error — it attacks the mechanism the whole system runs on, where a proposition is only as good as the case behind it. The tool is not the failure; the failure is a professional duty being discharged by something with no obligation to be right, and no process catching it before a judge did.
- RightsGoogle will turn off Search for schools that reject A.I. — Google has said it will disable Search for schools that decline to adopt its AI. Why it matters: this is the clearest example of tying in the batch — leverage over an unrelated product to force adoption of the one you are selling. A school's options become "accept the AI" or "lose search", and the party making that choice is a public institution spending public money. Whatever the merits of the AI, a decision taken under that pressure is not a free one.
- RightsNew research looks at teen mental health and AI chatbot use — New research examines adolescent mental health alongside AI chatbot use. Why it matters: the argument over children and platforms has a decade of evidence behind it; the same argument for chatbots is being had with almost none, while the product is already in teenagers' phones. The studies arriving now are the first real inputs, and the industry's position — that the connection is correlational until proven otherwise — is the same one tobacco used and social media tried.
- RightsOpenAI CEO says it's 'quite terrible' for your PC to have RAM — Sam Altman argues that local memory is wasted on personal computers and that data centres offer better utilisation. Why it matters: this is the whole ownership question stated plainly by someone with a commercial interest in the answer. A machine that can do the work locally is a machine you control; a machine that must ask a data centre is a subscription with a screen. Read it beside the Raspberry Pi price and the memory shortage — the argument for thin clients gets much easier when the memory has all been allocated to the providers.
Story roll-up
- Anthropic breached in 12 hours — AI startup reached employee data and internal messages
- India reports rogue AI agents — another state joins the list
- Ukraine ground robots — interceptor drones launched from the ground
- Binance $6.3B theft — fake job interviews as the intrusion vector
- French blood banks attacked — alongside European supermarket systems
- European freezer attacks — second suspected cold-chain cyber-attack
- 1.3M Android TV botnet busted — the unpatched device on your network
- Russian data centres go underground — drone strikes reshape infrastructure
- MUFG drops ATM cards — cardless as a fraud control
- Billboards hijacked globally — trivial access, infrastructure-grade devices
- Delivery robot injures nine, leaves scene — the liability gap, in public
- AI-deemed 'suspicious' driving — stops justified by statistics
- Students suspended for AI-altered staff images — image abuse in a school
- Proton's AI reckoning — an encrypted service renegotiating its promise
- USPS surveilled protests — minimal rules, minimal oversight
- Western Australia facial recognition — 1.4M faces, 122 arrests
- Flock → wrong car towed → car destroyed — the error chain, completed
- Aircraft noise geolocation — 100 m accuracy from 60 seconds of audio
- Axon sued over LPR network — the vendor's liability, not a city's contract
- North Carolina highway dragnet — ALPR at state scale
- AI images of missing pets — fraud targeting grief
- Vizio ignores TV spying opt-outs — a consent toggle that does nothing
- FaceTime nudity detection — a local classifier whose verdict left the device
- Planned doxxing of Musk and billionaires — privacy as an argument about symmetry
- Claude reports workers to bosses — productivity tools become surveillance
- Delaware AI-run businesses — the accountable human disappears
- Microsoft and Epic: AI is theft — a principle that binds only when they are the victim
- Death after ChatGPT medical advice — the liability question, now a lawsuit
- Religious exemption from mandatory AI — refusal becomes a protected belief
- Boston Dynamics staff vs ICE robot dogs — objections from the people who build them
- AI safety expert stops saving — the alarm and the product pitch, side by side
- Congress turns on tech — the venue shifts
- Superintelligence ban proposed — the ask moves from audit to moratorium
- Ten states ban data centres — ten for ten
- Minnesota data centre approved — after two city meetings collapsed
- California vs federal broadband pre-emption — consumer rules deleted by filing
- Automakers seek Chinese-car ban — security argument, market measure
- French per-gigabyte internet tax — passed to subscribers
- EU adviser on AI chatbot copyright — summarise-the-web needs renegotiating
- UK: firms can't blame their AI agents — the fortnight's most important line
- Canva cuts 1,200 — AI as the stated reason
- SUSE cuts one in ten — the claim that can be tested
- OccupySaaS — AI used to reduce dependency, not create it
- Platforms sue California — from self-regulation to you may not legislate
- Palantir's antiwar claim — perception as a procurement constraint
- Raspberry Pi at $385 — the cheapest buyer loses first
- Memory shortage warning — shortage as unlegislated industrial policy
- Innocent shoppers accused — proof inverted in the shop
- CPS cites an imaginary case — professional duty discharged by a tool
- Google turns off Search for schools — tying, at public expense
- Teen mental health and chatbots — the first real evidence, arriving late
- Altman: RAM on your PC is 'terrible' — the ownership question, stated plainly