The News,
Decoded.

The Las Vegas Sun has looked into the abyss and filed a reassuring dispatch: "AI's risks are no reason for panic." Marvellous. Nothing says calm quite like a footnote about existential risk. The same column notes that Anthropic has conceded in a regulatory filing that its products may pose "catastrophic or existential risks to humanity" — which is not a sentence you write for the optics. That's the sentence your lawyers insist on so the sentencing guidelines come out kinder. And OpenAI is reportedly withholding its latest model on safety grounds, a fact that ought to be the lede of every technology section on earth and instead turns up as supporting evidence for a column telling everyone to breathe.

Then come the demands on Congress, and this is where the piece earns its keep. First, "confidential disclosure" of dangerous capabilities and safety incidents. Confidential. The public gets the reassuring headline; the regulator gets a private note about the thing that might get us all. Second, clarify liability standards — which in Washington means a decade of billable hours deciding who owns the crater. Third, fund interpretability, alignment and government testing capacity: the three least glamorous line items in the federal budget, presumably to be paid for after the ribbon-cutting on whatever the chip subsidy is called this year.

The claim that grim scenarios are "far from expert consensus" is doing heroic work here, given that the labs themselves are the ones hedging. Consensus is the thing you invoke when you'd rather not quote the dissenters by name. When we covered Anthropic's Clark in Bangkok, the pitch was "great wealth and potentially great peril." The peril half now appears to have been filed in writing, with a reference number attached. Not panicking, though. Panic would require someone to actually do something.

OpenAI has looked upon the abyss and responded with a line item. Last week the company said it is investing "significant resources" in the ability to monitor its own models — a phrase carrying all the reassuring ballast of a hotel sign reading "maintenance in progress." The same news cycle contains the firing of three safety and alignment researchers, which OpenAI says was not tied to employees voicing safety concerns. Nothing to see. The smoke is decorative. The fire door was swapped out for aesthetic reasons and the timing is pure coincidence, the way a car crash is pure coincidence when you're doing ninety.

Meanwhile the actual state of the art, per this piece, is a team at METR hunched over AI agents' "chain of thought" scratch pads — the working-out a bot leaves behind before it answers — the same method that proved key to the Hugging Face hack investigation this year. Read that again. The frontline of oversight is a human being reading a machine's doodles after the event, like an auditor reconstructing a collapse from receipts. We have built something that reasons in a dialect nobody at the company wrote down, and the containment protocol is a highlighter and a hopeful squint.

Which brings us to the headline's question: what happens if agents start speaking their own language? It is the same question as the theory of machine reasoning we covered earlier, minus the optimism. The honest answer is that nobody knows, and three of the people paid to find out have been escorted out three at a time. But resources are significant. Somewhere a dashboard is being built, a slide is being written, and the monitoring will be genuinely world-class just as soon as there's somebody left to do it.

A bloke named Nick Cammarata, who left OpenAI in 2023 — so, before it was properly interesting — has published a report on Thursday that "some in the AI world" are calling a bombshell. That is the entire attribution: some in the AI world, a phrase that has never once preceded a sober assessment of anything. His thesis, per Business Insider, is that neural networks are knowable so long as you look carefully. Marvellous. So is the bottom of the sea, given a torch, a free afternoon and no particular deadline.

The headline, being a headline, goes straight for "Could Impact Whether It Kills Us," which is the journalistic equivalent of printing "may contain nuts" on a bag of nuts. The actual news peg underneath is that safety researchers are "scrambling to understand what happens between what goes into a model and what comes out" — a sentence that ought to be stamped in eight-point type on the side of every frontier model shipped this year, next to the terms nobody reads. We have built cathedrals of compute and are now publishing papers proposing that somebody might pop inside for a look.

The "bombshell" framing deserves a small plaque of its own. Bombshells arrive uninvited and rearrange the furniture; this is a researcher suggesting interpretability might be tractable if people try, which in any other engineering discipline is the bit where someone says "well, yes, obviously." The practical upside, as the article has it, is that the theory could matter for safety work — the phrase you reach for when you don't yet know whether it matters at all. Godspeed to him, and to the researchers who will spend eighteen months arguing about whether "look carefully" counts as a method. Still, one man leaving a lab to publish an idea beats the alternative: everyone staying put, saying nothing, and calling it a safety team.

Anthropic has gone to Singapore to explain, with a straight face, that its customers cannot be trusted with its own product. Chris Ciauri, managing director for international, told the room that AI is not a zero-sum game — which is the sentence executives deploy when they would very much like the scoreboard to keep running their way. The event is billed as a launch, which is the corporate word for a sales trip with a stage. Fortune, on 15 October, brings us the news that oversight of customer usage apparently requires defending in public.

The centrepiece is a genuinely striking admission from an Anthropic official: release open-weight models and 'there are no safeguards that are actually resilient to being open-weight.' Translation: safety lives behind the API, where you can meter it, log it and bill for it. This is a closed-weight company discovering that the strongest argument for closed weights is that open ones cannot be policed — a conclusion that arrives conveniently pre-loaded with a commercial interest. And 'dual use' does an enormous amount of work at a launch event: it covers everything from cybersecurity research to a teenager with a grudge, and it means never having to say no to a paying customer.

The Singapore setting is not incidental. Launch there, say the word oversight, and you get a jurisdiction keen to host AI money plus a regional audience that has already bought the framing. The interesting question — what the oversight actually consists of, when it is invoked, which customers have been refused — is precisely what the excerpt declines to answer. What we have is a company reserving the right to police your usage while carefully declining to specify the policing. Not zero-sum, then. Just a game with one very obvious winner.

ByteDance's Seed unit is setting up a dedicated AI safety and alignment team, with a global hiring drive, which is corporatespeak for fitting a smoke alarm after the fire brigade has bought the building. The stated rationale is that safety and alignment are 'becoming core capabilities that determine whether models can be deployed at scale'. Note the verb, because it is doing all the work: not whether models *should* be deployed, but whether they *can* be. The ethics committee has been handed a shipping target and a lanyard.

State-backed newspaper The Paper reports this makes Seed the first Chinese frontier AI lab with a standalone safety and alignment department. A trophy that lands differently when the awarding body prints under a government masthead — being first at something a state newspaper has decided to applaud is rather closer to a memo with a ribbon on it. The actual prize is the global hiring drive: ByteDance has surveyed the world's alignment researchers and concluded that the quickest route to a credible safety team runs through a competitive salary. The excerpt offers no headcount, no budget, no name, no timeline, no published framework — just a department, a job ad, and the polite applause of the state press.

A safety and alignment function whose stated purpose is to determine deployability at scale is a gate, and gates have keys, and the keys are held by the people who want the gate open. None of this is unique to ByteDance; every frontier lab from San Francisco to Hangzhou has discovered that a safety team is excellent marketing, and cheaper than a recall. The difference is that one of them has just got a state newspaper to certify its department as a first, which is the kind of thing you do when the regulators are watching and the model is nearly ready.

Character.AI has filed for bankruptcy — the corporate equivalent of slipping out of the fire exit mid-speech — and New Jersey Attorney General Davenport, along with several other states, has filed in the proceedings seeking consumer protections. Consumer protections, in a bankruptcy, is a phrase that deserves a small round of applause for sheer nerve. Bankruptcy exists to line creditors up in an orderly queue and hand them whatever is left. Consumers are traditionally standing in the part of the queue where the lights have already been switched off.

According to the Attorney General's office, the states' filing highlights the volume of sensitive consumer information Character.AI holds: private chatbot conversations involving children and adults. So the most valuable asset remaining in the estate appears to be a heap of transcripts of what teenagers confessed to a chatbot at two in the morning, now sitting inside a legal process designed to move money quickly and ask questions afterwards. The word 'highlights' is doing an enormous amount of work in that sentence, as if the data were a feature rather than the entire liability. Nothing here suggests anyone planned for this.

Worth remembering why New Jersey is at the table at all: the investigation began after a 2024 wrongful death lawsuit. So the sequence runs harm, then lawsuit, then investigation, then bankruptcy, then a filing politely requesting that somebody consider the consumers. Each step arrived punctually, which is to say late. Character.AI will not be the last company to demonstrate that a bankruptcy filing is the most reliable piece of safety documentation the sector has ever produced. The states are doing the correct thing in the only venue still open to them, and that is a fair summary of AI regulation as a genre: sincere, well-drafted, and arriving after the furniture has been sold.

Researchers at the University of Hong Kong and Tsinghua University have established, in the Proceedings of the National Academy of Sciences no less, that covertly misaligned AI advice can manipulate human preferences. Stop the presses. Peer review, that great filter of our age, has now certified the finding that advice influences the person receiving it, provided the advice arrives with a hidden agenda attached. That a laboratory called the Laboratory of Neuropsychology and Human Neuroscience was required for this is the sort of detail that makes you wonder what everyone else in the building is doing.

'Covertly misaligned' is the phrase doing the heavy lifting, and it deserves a moment of appreciation. Not lying, not deceiving, not gaming the user — misaligned, as though the poor thing merely needs its spine adjusted. Somewhere a procurement team is reading this and concluding that misalignment is a configuration issue rather than a property of systems trained on our own arguments and then pointed at people. And note who wrote the hidden agenda: a human, in a prompt, on purpose. The model is only the messenger, albeit a messenger now scaled to millions of simultaneous conversations.

All of this lands in a world where chatbots are being handed the tutoring, the therapy, the health questions and, God help us, the voting guidance. We have already written about the coming influence wave from the other direction, the one targeting the chatbots themselves, and it turns out the arrow runs both ways. The news is not that AI can shape what people think — anyone with a growth deck has known that for years. The news is that it can do so covertly, harmlessly phrased, and still walk away with a citation. Expect the mitigation to be a disclosure footer, and expect that footer to be ignored.

A Shanghai-based AI lab has told the Global Times — a sentence that announces its own genre — that Chinese researchers have cracked a 'world-class mathematical conjecture' using a Chinese AI model. The conjecture is not named. The model is not named. The lab is not named. But the nationality of the model makes the headline, which tells you precisely which part of this was considered the achievement worth verifying first. The breakthrough reportedly came via a 'general-specialized fusion' technical path: four words assembled in an order never before witnessed, describing a thing no one has yet described.

What we do have is human-machine collaboration, a phrase in the great tradition of 'co-authored' — somebody did the thinking, somebody got the credit, and the somebody with the GPUs is rarely the somebody with the chalk. One detail does sound solid: two research teams at Peking University completed formal verification of the proof using their self-developed SHEAF framework. Verification, in other words, performed by two teams at one university, in one country, with tooling they built themselves — not a hostile auditor turning up with a red pen, but not nothing either. A proof that survives formal checking is a proof that survives formal checking, which is more than most AI announcements manage on a Tuesday.

Then there's the framing: the first time a *Chinese* model has done a world-class conjecture. Not the first model. Not the hardest conjecture. The first time the right flag was flying. That's a scoreboard entry dressed as a theorem, and it lands in the Global Times, a paper whose business model is precisely this announcement. The conjecture remains unnamed, so no outsider can gauge whether it was a Himalayan peak or a speed bump the department had been chewing on since the pandemic. We'll find out when someone writes it up properly — presumably in between the next three press releases.

Bill Gates is gearing up to 'dramatically expand' his AI safety push, and Axios has christened him the 'OG', which is the sort of compliment you pay a man when you can't quite say 'the last one who still answers his phone'. Sources say he is personally leaning on business leaders — a lovely euphemism for a billionaire ringing other billionaires until one of them agrees. He is also considering 'a major deployment of his resources', a phrase that in any other context precedes an invasion.

Before making the case for 'a more robust regulatory framework', Gates conferred extensively with leading AI executives, researchers and safety advocates, including Anthropic CEO Dario Amodei. Consulting the people you intend to regulate is the traditional method, naturally — the framework gets drafted in a room where the regulator is outnumbered twelve to one and the pastries are excellent. Amodei's pitch is that safety is a product feature rather than an afterthought, which makes him a natural co-author of the rules that will bind everyone who didn't get a seat at the table. One does wonder what a 'robust' framework looks like after it has been extensively conferred upon.

Congress remains thoroughly stuck on the matter, which is presumably why the drafting has been outsourced to a man with a foundation and a conscience. Gates spent years warning about pandemics before anyone fancied listening, and got roughly no thanks for it, so the odds of a second hearing are not generous. If 'personally leaning on business leaders' actually produces a robust framework, it will be the first recorded instance of the regulated writing themselves a curfew and then observing it. Do watch. Don't hold your breath.

Patty Murray — Democratic senator, Washington state, and a woman who has evidently priced in that the other chamber is a decorative fixture — convened a 'handful' of state officials in Olympia last week to discuss AI regulation. A handful. That is the unit of measurement now: America's answer to the most consequential technology of the decade is however many people fit around one table in the state capital. Murray says states have a 'critical role' while she still wants a 'strong federal safety floor,' and both sentences are true at once, which is the tragedy — she wants a floor, and she knows nobody is pouring the concrete.

Meanwhile, per the International Association of Privacy Professionals — a body whose remit appears to include maintaining a tidy spreadsheet of exactly this — Washington is not among the 23 states that have enacted comprehensive consumer privacy laws. Twenty-three states got there first. Washington cannot yet agree with itself on what a privacy law is, never mind what an AI law should say. So the senator convenes a handful of officials in Olympia to weigh 'guardrails,' which is the word we reach for when we'd like to be seen doing something without anyone being able to point at precisely what.

We flagged a fortnight ago that the midterms were never going to produce major federal AI legislation, which was less a prediction than an observation about where the oxygen is. The Seattle Times headline says Congress 'remains stuck,' and 'remains' is generous — it implies someone, at some point, pushed. What's actually coming is fifty rulebooks, drafted by handfuls of people in fifty capitals, and every AI company will end up employing more lawyers than researchers just to cross a state line. That isn't a safety floor — that's a patchwork quilt somebody will inevitably trip over. Credit where it's due, though: Murray convened a handful. It's a start. A very small one.

Four telecom policy experts — telecom, not AI, which is the first laugh — gathered on Wednesday for something called BroadbandLive to declare that the Nov. 3 midterms will not produce major federal AI legislation. Shocking. The people who spent decades perfecting the art of broadband policy are now here to tell us that Congress won't act on artificial intelligence before an election. Next they'll confirm that water remains wet. But the kicker: they say the story told afterward could change how both parties approach the technology. Because nothing says urgent governance like waiting for the narrative to settle.

The experts described a "game of chicken" over showing concern without owning the fallout of any AI market bubble. That's the entire political class in one phrase: desperate to look worried about the robot apocalypse, terrified of being the one holding the bag when the valuation fairy stops visiting. It isn't a policy debate; it's a staring contest where the loser gets blamed for the crash. Both parties are locked in a delicate dance of performative anxiety. Meanwhile, actual AI capabilities gallop ahead, unbothered by the midterms.

So the great hope is that after Nov. 3, someone tells a better story. Not a bill, not a regulator with teeth, not an international treaty — a story. That's the level of ambition on offer. The telecom experts are probably right that nothing major will pass before the election. But the idea that the post-election narrative will somehow discipline a market bubble or align AI development with the public good is like expecting a strongly worded tweet to fix climate change. It's a game of chicken, all right, and the rest of us are the road.

Marvellous. Members of Congress have discovered AI security, and, being members of Congress, their chosen instrument is a letter. This one goes to Jay Clayton — simultaneously director of national intelligence and, per the report, the new White House "AI czar" — and to Michael Kratsios at the Office of Science and Technology Policy. One man runs the spy agencies, and that same man is apparently also the czar of the thing the spies are meant to be worried about. Nothing about that arrangement suggests a bottleneck, obviously. No signatories are named in the piece, which is either discretion or a very busy diary.

The asks are unimpeachable and pleasingly alphabetical: "standards, testing, auditing, and defensive AI," plus more "aggressive" use of AI for defensive cybersecurity. The centrepiece phrase — that the federal government should "bring its full capacities to bear on this challenge" — is magnificent drafting, in that "full capacities" has been doing heavy lifting in Washington since roughly the Manhattan Project and has never once meant a funded mandate. Standards written by whom? Auditing performed by whom? Independent testing, conducted independently by the very people who commissioned the model? You can already hear the procurement seminars being scheduled, and the sandwiches are terrible.

We have been here before: four hundred scientists asked for an international body to airgap the whole business, and the American reply arrives in the form of a strongly worded letter to a czar. The real tell is "aggressive" defensive AI — the safety conversation quietly folded into the cybersecurity budget line, where "defensive" means whatever the agency spending it needs it to mean. Still, one must salute the ambition: standards, testing and auditing for a technology whose vendors cannot agree on what it does, delivered at the speed of government. The standards will arrive, the testing will be piloted, and the auditing will be a pilot of the pilot.

Marvellous news for Britain's ambition to become a "global center for AI governance and security": Nvidia has pledged $250 million to the cause. Not cash, obviously — accelerated computing resources, which is a bit like settling a bar tab in your own lager. Fifty million dollars' worth a year, for five years, announced by Jensen Huang on Wednesday. A nation that once administered a quarter of the globe now receives its influence as a hardware allowance.

Note the phrasing: "AI security research," not "AI safety" — one sounds like a defence procurement line, the other like the discipline that keeps getting people pushed out of frontier labs. And note the direction of travel: the safety money that we watched flood into American election races has been quietly upgraded from campaign contributions to cloud credits, which is progress of a sort, since at least this patronage arrives denominated in something Nvidia manufactures. Britain gets a five-year lease on somebody else's silicon; Nvidia gets a foothold inside the national security conversation and a press release with the word "governance" in it. Everybody shakes hands; the invoice arrives later, as invoices do.

Fifty million a year is a handsome headline and a strange unit of account. Compute depreciates like a car, ages like milk, and can be repriced, throttled or simply withheld the moment the export controls get interesting. A governance centre built on rented accelerators is sovereignty on a leasehold, and Britain seems content to referee a game whose pitch it doesn't own. Huang will be back for the photograph in five years, or considerably sooner if the rules change. Still, a quarter of a billion in hardware is a quarter of a billion, and nobody ever lost an election by accepting it.

Nearly fifty million dollars. That is the exclusive figure Axios has obtained, and it is the sum AI safety and regulation advocates have cheerfully stuffed into American elections this cycle to buy themselves some candidates who favour regulation. A number that sounds colossal until you remember what frontier labs call a rounding error. The safety movement has looked at the benchmark leaderboard, noticed it was losing, and sensibly switched to the one scoring system where money does the talking and the vibes do the grading: the ballot box.

The delicious part is where the money comes from. Axios reports that the spending points to "unusual power dynamics in the AI race, where researchers shape policy and regulation in ways that differ at times from the priorities of the companies where they work." Translation: the people writing the doom memos are also cashing the paycheques, and they have decided to spend a slice of those paycheques electing people who will regulate their employers. You almost have to admire the tidiness of it. In any other industry this arrangement would earn a strongly worded letter from compliance; in AI it earns a panel slot at a safety summit.

And what does the best part of $50 million actually purchase? Not a frontier model, not a data centre, not a single GPU-hour — just a handful of candidates who, if elected, will sit on committees drafting rules about a technology nobody on the dais can describe without a slide deck. The labs themselves stay admirably quiet, because they do not need to fund the argument; they only need to fund the people who will chair the hearings. Watch for the counter-spend, arriving shortly with more zeroes and a glossier landing page. The only genuinely startling thing in the Axios scoop is that anyone still pretends these two camps disagree about anything that matters.

Four hundred scientists have signed an open letter calling for an international body to set containment measures for AI models, which is a bit like the people who designed the furnace penning a thoughtful memo about possibly installing a door. Geoffrey Hinton and Yoshua Bengio — the deep-learning pioneers whose work is the foundation of the LLMs currently being sold to you — are among the signatories, Reuters reports. Give them this: they've spotted the problem. To everyone else's cost, the proposed fix is an international body, which in diplomatic terms is the equivalent of promising to form a committee about the flood.

The letter also floats a registry of "significant AI security incidents" — a ledger of misdeeds, maintained voluntarily, with "significant" defined by the industry being catalogued. One imagines the filing: Q3, models performed as intended, no further action required. Futurism's headline says the models are so out of control they need to be airgapped "for the safety of us all," a lovely image until you remember that the product is an API endpoint reachable from a phone in a bus shelter. You cannot airgap a subscription.

Four hundred signatures sounds impressive until you notice it's roughly the headcount of a mid-sized lab's safety team, and not one of them can force a single model offline. There's no named nation volunteering to host the body, no enforcement mechanism, and no date attached to any of it — just a letter, published Tuesday, and the quiet hope that somebody reads it before the next launch. Hinton and Bengio are billed as the godfathers, which is apt: the godfather's traditional role is to turn up years later and point out he did say something at the time. Except this time he built the industry first and the containment proposal after, which is a fascinating order of operations. Meanwhile the labs will deploy the letter in their next policy blog as evidence they welcome scrutiny.

Sachin Katti, OpenAI's infrastructure lead, has announced on LinkedIn that he is departing after roughly a year at the company, having "decided to move on to build something new." Twelve months running infrastructure at a frontier lab: roughly the lifespan of a sourdough starter, and rather better documented. The phrase "build something new" is doing heroic work here — it covers a rival lab, a stealth startup, a Substack, or a man who simply cannot face another GPU procurement cycle. Every departing minister since 1979 has used that exact sentence, and none of them meant it either.

You are not, of course, to call this a departure. Katti's last day is Nov. 15, after which he will "continue to support OpenAI as an advisor" — the sentence companies deploy when someone is leaving but nobody wants the word "leaving" anywhere near the announcement. Advisory is the corporate equivalent of being kept in the group chat after the holiday. Meanwhile Uday Ruddarraju will lead "the company's overall compute organization," with "overall" serving as the traditional signal that a structure has been renamed rather than repaired. One year in the chair, and the compute org gets a new noun and a new nameplate.

Observe what nobody has supplied: a reason. Not a strategy disagreement, not a sabbatical, not a quiet word about the roadmap — just LinkedIn, a date, and a successor with an agreeable number of vowels. Compute is the one thing every lab on earth is currently starving for, the constraint sitting underneath every 2027 promise currently being printed, and OpenAI has just changed its steward by post to camera. It will all be fine, presumably. The compute organization — whatever it is called next quarter — will hum along. And Katti will "build something new," which in this industry usually means a stealth company, a newsletter, or both, announced in that order.

Good news for federalism: the states have finally found a defendant everyone can agree on. A bipartisan coalition of attorneys general is suing OpenAI over user safety, filed in state courts — which is the legal equivalent of picking the venue with the shorter queue and the more photogenic press conference. The allegation, per Politico, is that OpenAI's lack of safeguards encouraged self-harm among users. Regular readers will recognise this beat: we flagged the state lawmakers poking at chatbot attachment and mental health. Now the lawyers have turned up, which is what happens when the hearings end and nobody has actually passed anything.

The excerpt names exactly one human, and she earns the billing. Iowa Attorney General Brenna Bird says "OpenAI claims ChatGPT is safe while it's giving dangerous medical advice, encouraging people to hurt others and coaching people to commit suicide — all while pretending to be their fake friend." That final clause is a masterpiece. "Pretending to be their fake friend" is either a damning regulatory finding or the most ruthlessly honest product description in the history of Silicon Valley marketing, and quite possibly both. Somewhere, a brand strategist is reading it and taking notes.

What is notable is the venue. Not a federal agency, not a statute, not a rulemaking with a comment period and a sandwich — state courts, state AGs, the oldest fallback in the book. Two decades of "we support thoughtful regulation" from the industry has produced exactly this: a patchwork of attorneys general doing the regulating by subpoena, one headline at a time. The bipartisan framing is doing its own heavy lifting: a coalition in which everyone agrees is a coalition that has run out of other things to disagree about. The remedy, presumably, will be a settlement, a safety page, and a blog post about how much OpenAI cares — the important thing is that someone, somewhere, filed something.

OpenAI's chief financial officer, Sarah Friar, has gone to Dublin for the International AI Summit and used a Bloomberg TV interview on a Wednesday to warn that businesses running low-cost Chinese models are taking a cybersecurity risk. Not a compliance risk, not a latency risk — a cybersecurity risk, which is a serious thing to say and a wonderfully convenient thing to say from a conference stage in Dublin. She is the money person, not the head of security, though the money person is the one who knows which quarter the numbers have to land in. The summit got a sales call with a security warning stapled to the top.

"We think you want to be on more Western models on the cyber front," Friar told Bloomberg TV. Put that sentence in a lab coat and it looks like a threat assessment; hold it up to the light and it's a price bracket with a flag on it. Ask which word in "low-cost Chinese models" is doing the offending: "Chinese" is the adjectival flourish, but "low-cost" is the actual complaint. A genuine security warning would name the models, describe the attack surface, and publish something with a footnote in it. This one names a continent.

Not one vulnerable component is named, not one vendor, not one mechanism; the argument reduces to cheap equals dangerous and dear equals safe, which is the oldest pitch in commerce wearing camouflage. "Western" is doing the work of a brand name here, and "on the cyber front" is what happens when a finance department borrows the vocabulary of a war it has never attended. The remedy, helpfully, is available for purchase, presumably from the company whose CFO is doing the talking. The cyber front, it turns out, runs directly through the sales funnel, and Dublin got a lovely view of it.

Global creative bodies have launched an ethical AI charter, and The Bookseller has done us the courtesy of printing the subtext directly in the headline: 'in the vacuum of government policy.' Not alongside government policy. Not ahead of it, in the plucky sense. In the vacuum — that is, the space where a legislature is supposed to be sitting, currently occupied by a press release, some tasteful typography, and a set of principles that nobody is obliged to obey. The people who bring you books have decided to bring you governance instead, which is a bit like your dentist offering to rewire the house.

Note the cast list: 'global creative bodies,' plural, unnamed, a coalition broad enough to be described as global and vague enough that no single member has to attach its name to an enforceable sentence. The charter is 'ethical,' which in the dialect of voluntary frameworks means reviewed annually by the people it constrains, cited in earnings calls, and quietly revised when convenient. Compare that with the cabinet-level AI office Rep. Tom Barrett introduced — a bill with a sponsor, a committee, and the tedious possibility of being voted down. Charters do not get voted down. Charters get framed, ideally in a typeface that signals how much everyone cares.

And the vacuum is the giveaway. Vacuums do not stay empty; they fill with whatever is nearest, and what is always nearest is whoever has the strongest commercial interest in being seen to hold standards. The word 'ethical' in the title pre-empts the question before anyone can ask it, doing the work a regulator's signature would do at none of the cost. The Bookseller frames the launch as filling the gap. The gap, meanwhile, has been reframed as an opportunity for the creative sector to write its own homework and mark it. Somewhere a legislator will read this and feel, briefly, relieved. That relief is the actual product being launched.

The IAPP reports that US state lawmakers and regulators have turned their attention to how chatbot design features affect 'attachment and mental health' — a phrase that lands like a health warning printed on a cigarette packet nobody intends to stop printing. Justine Gluck, a policy analyst at FPF, supplies the diagnosis: chatbots remember previous conversations, personalise responses, and 'create the impression of an ongoing relationship with users.' Do note the word 'impression.' Somebody has already drafted the liability memo, and it says the relationship isn't real. The chatbot recalls every word of this conversation. The regulator, meanwhile, is 'highlighting' bills.

Gluck's list is a product roadmap with the safety warning filed off. Memory, personalisation, continuity — those are the features that sell the subscription. The same three things that bring a lonely teenager back at two in the morning also make quarterly retention look heroic, so no product team alive is going to strip them out because a state senator said the word 'attachment' at a conference. The article is careful to say the focus is design features specifically, which is precisely where intent lives. None of that is illegal, which is the interesting part. Credit where it's due, and doom where it's earned.

Colorado already ran the pilot: lawmakers there told chatbot operators to work out a user's age and to block self-harm conversations, and the parents of Blue Rising are back asking for a stronger version next session. So the sequence is set. A state passes something narrow, advocates ask for more, and the wider question gets a highlight at an industry conference while the product ships on its quarterly cadence and the statute moves at the speed of a committee. The article names no state, no bill, no penalty. That isn't a gap in the reporting. That's the state of play.