The News,
Decoded.

The Washington Post's AI & Tech Brief has an exclusive: safety groups reject the Thune-Klobuchar bill. Two names stapled together with a hyphen, and the answer was still a flat no. The brief does not trouble its readers with what the bill actually contains; it simply tells us who is against it, which in this town apparently constitutes the entire scoop. The rejection is the headline, and the substance is a footnote nobody wrote.

What the safety crowd wants instead: embedded evaluators inside the frontier AI labs. Sit with that for a second. The ask is to be seated within the entities you are meant to scrutinise — a lanyard, a hot desk, and a badge that opens every door except the rooms where the models actually get built. We flagged this exact shape of gap when three states legislated incident reporting and left the forensic follow-through off the invoice. Reporting duties are what you draft when you have quietly decided you would rather not know, and embedding the referee inside the fox's kitchen carries precisely the same aroma.

Meanwhile the Thune-Klobuchar bill joins the pile, and the wider signal out of Washington has been that no new AI law is likely to reach a signature anyway — which makes the rejection faintly hilarious, the political equivalent of returning a parcel to a warehouse that burned down last spring. There is a genuinely bleak comedy in watching the most worried people in the room decline the only seat offered because the seat comes with the blindfold already fitted. Two names get the letterhead. The evaluators get a lanyard and a hot desk. Everyone else gets an exclusive and another bill nobody will ever have to comply with.

Three American states have now legislated on AI safety — California's SB 53, New York's Responsible AI Safety and Education Act, and Illinois's SB 315 — and Lawfare has found the detail everyone else stepped over: the laws require frontier developers to report certain safety incidents to state authorities, and nobody has built the forensics to check a single one of those reports. So the enforceable architecture of AI safety across three jurisdictions currently runs on the honour system. Which is a curious choice for an industry whose entire pitch is that it is building the most dangerous technology in human history while remaining the only entity competent to supervise it.

The revenue threshold tells you who the laws are actually for. The reporting, transparency and independent audit duties bite only above $500 million in gross annual revenue, which is a tidy way of describing the handful of labs that can retain the lawyers to argue about what counts as a reportable incident in the first place. Meanwhile Congress, as we noted, is not about to pass anything the President would sign, so the federal floor stays a rumour while three states invent three separate vocabularies for the same failure modes.

And the forensic gap is the punchline. A report is a document. Absent chain of custody, a standard taxonomy of incidents, logs anyone can subpoena, or one investigator with technical training, what a state authority receives is a filing drafted by the lab that caused the incident, describing the incident in the lab's preferred tense. It is asking the fox to complete the henhouse incident form in triplicate and then complimenting the handwriting. Three statutes, one revenue cut-off, zero forensic capacity — the most comfortable safety regime ever designed, and the labs didn't even have to write it themselves.

The UK and Canada have held their first bilateral meeting, and the agenda included AI risks — a subject on which neither country has actually legislated, which does save a great deal of drafting time. Canada's Artificial Intelligence and Data Act lapsed when parliament was prorogued in January 2025, which is parliamentary procedure for 'the bill died because everyone went home.' The UK has been running a pro-innovation, sector-led approach since its policy paper in March 2023, and the approach has been running ever since, unbothered by outcomes. So two governments with no functioning AI statute have convened to discuss the risks posed by AI. It is two bald men, one comb.

The timing is a small masterpiece. Britain's sector-led arrangement means the sector leads, and the sector's settled preference is that nobody leads at all — a governance model with the enormous advantage of being invisible to voters, regulators and anyone with a subpoena. So Burnham and Carney met, discussed AI risks, and produced a photograph, which in fairness is more than the AIDA managed. The risks, at minimum, were discussed.

The substantive item is 'defence cooperation,' which is where AI stops being a values seminar and becomes procurement. Defence is the one AI domain where nobody proposes a three-year consultation, because the counterparty is not consulting. Meanwhile the capability calendar is set in labs, and the legislative calendar in London and Ottawa is set by whether anyone fancies a recess — a mismatch that the first bilateral meeting did not so much address as illustrate. Still, a milestone: two countries with no dedicated AI law have now agreed to keep discussing AI. The bill lapsed. The conversation continues. One of those is cheaper.

Trump is all-in on AI, per the BBC, which reads like a strategy until you remember this is the same political operation that gave the world the 'AI hoax' line. Congressional legislation he'd actually sign? Unlikely — not because the bill would be bad, but because presidential resistance means it never gets that far. Mike Johnson, previously open to congressional action, has now backed away and echoed the hoax rhetoric instead. Nothing says winning like handing the rulebook to whoever ships fastest and calling it deregulation.

The warnings, presumably, are decorative. The BBC frames the story as Trump going all-in 'despite' them, a preposition doing heroic structural work — 'despite' implies somebody read the warnings before filing them under later. We flagged this exact pattern when a key lawmaker suggested AI safety legislation could simply wait; now the Speaker of the House is running the same play at altitude. Congress isn't gridlocked, it's ornamental. Nobody in this story is pretending the warnings were wrong; they're simply not load-bearing. The labs get the runway, the voters get the vibes, and the warnings get a footnote.

So the most consequential technology of the century will be governed by a President who has called it a hoax, a Speaker who has decided to agree with him, and a legislative calendar with a hole where the bill should be. Somewhere a compliance officer is drafting a risk register that reads 'none' and clocking off early. The upside, if you squint: no regulatory overreach, because you'd need a regulation first. The downside is everything else — safety researchers publishing concerns into a vacuum, states assembling a patchwork, and the next capability jump landing before anyone writes up the last one. Trump going all-in means everyone else is all out of luck, and the warnings the BBC mentions will keep being mentioned, right up until the moment they're not.

Emil Michael, the Pentagon's chief technology officer, has surveyed the AI industry and arrived at the boldest available policy position for a man with a security clearance and a LinkedIn: the government should own none of it and regulate none of it. Taking equity stakes in AI companies "doesn't really go well in terms of nationalization," he told CNBC — a sentence that treats nationalisation as weather, something that happens to you rather than something a government decides to do. Somewhere, a sovereign wealth fund felt quietly judged. This is the man whose job involves worrying about Pentagon dependency on a handful of vendors, and his answer is: don't interfere.

He also "repeatedly signaled opposition" to ramping up government regulatory oversight of AI companies — signaling being Washington's favourite verb, meaning he definitely said it but would rather the transcript stayed vague. Then the flourish: it would be "dangerous to place all trust in the hands of a few AI chiefs." Correct! Genuinely correct! And yet the arrangement left standing is: no equity, no oversight, and a handful of AI chiefs holding the keys while we all agree to trust them. Rep. Brett Guthrie has already suggested the FRONTIER Act can wait — Washington's regulatory posture now has the structural integrity of a nice idea floated at a conference.

Worth noting what Emil Michael did not say, because the excerpt is a museum of absences: no company, no stake, no deal, no bill, no figure. Just a Pentagon CTO explaining that the state should keep its hands off both the cap table and the rulebook, because look what happened the last time anyone mentioned nationalization. That leaves the trust question exactly where Washington prefers it — with the AI chiefs, who have never once needed a rule they didn't write. The oversight will arrive eventually, presumably in the form of a strongly worded hearing about the strongly worded hearing that never happened. Regulate nothing, own nothing, and hope the handful of chiefs are feeling civic. What could go wrong that hasn't already been priced in?

The Albanese government is "considering" reforms — a verb that in Canberra means we have already decided but would like the news cycle to do the shouting — that would hand US big tech "unrestricted access" to Australian content. Unrestricted is the tell. Not licensed, not negotiated, not metered. Unrestricted. Everything Australians put online, scraped for AI, with the Guardian's headline asking, with admirable politeness, "How will you be affected?" Reader, you will be affected the way a hedge is affected by a strimmer. And the phrase "proposed new rules" is doing the heavy lifting of a government that would much rather you picture a whiteboard than a statute book.

Note the beneficiary. Not Australian labs, not the local research sector, not some plucky Melbourne startup — "US big tech", the same firms who have spent a decade explaining that they cannot possibly pay for the journalism they surface. Now they get the whole country as a training corpus, free at the point of extraction. The word "reform" is doing a lot of lifting there, in the way "reform" always does when the thing being reformed is your privacy and the thing benefiting is somebody's quarterly filing.

And the timing is immaculate. Washington is busy debating frontier safety legislation; New York is drafting rules before its first law has even taken effect; and Australia's contribution to the global conversation is a proposal to let foreign companies hoover up the Australian internet. The consultation, if there is one, will be warm, well-mannered, and scheduled for a time and place engineered for low attendance. The opt-out, if there is one, will be three menus deep and described as a "preference". The AI will be trained regardless, and "how will you be affected?" will be answered for you, in the past tense, by a model that has already read your post.

Two governments have pooled their pocket change and come up with up to $300 million for LawZero, a lab building something called Scientist AI — 'a fundamentally new form of AI designed to be trustworthy and safe from inception.' Safe from inception. Which does rather invite the question of what the rest of the industry has been doing from inception, and whether 'we'll bolt safety on later' was ever really a plan so much as a launch schedule. The name suggests a clean slate, which is generous, given this industry's slate has not been clean since roughly the first foundation model.

The money comes from Canada and Germany, jointly, and the operative phrase is 'up to $300 million' — a commitment structured with the same escape hatch as a gym membership sold in January. It is also the sort of sum that, in a sector where a single lab's rumoured funding round gets discussed in the plural trillions, buys roughly one well-attended conference and a press release with a flag on it. Meanwhile in Washington, AI safety legislation has been granted the gift of time as we noted recently — no vote, no deadline, no bother.

And the technical pitch is a quiet indictment of everyone else: Scientist AI 'reasons transparently and does not pursue its own goals.' Imagine that. A model that reasons transparently, in an industry that has spent five years treating 'what's going on inside' as proprietary, and that does not pursue its own goals, in an industry where every assistant is described as pursuing yours while quietly pursuing its own. The transparency is the radical part; the not-pursuing-its-own-goals clause is the part that should embarrass every product page currently promising the opposite in a nicer font. Two governments, up to $300 million, for a system that doesn't lie about its motives. Progress, of a sort, at a price.

Rep. Brett Guthrie has gazed upon the FRONTIER Act — a piece of AI safety legislation backed by OpenAI, Anthropic, "important policy groups" and a growing number of federal lawmakers from both parties — and produced the most Washington verb in the dictionary: wait. Not no, not never — just wait, the word that keeps every option open and no one's name signed to anything. Rep. Jay Obernolte wanted a committee vote in November; Guthrie, described as a key lawmaker, appears to oppose it. November, incidentally, is a month that exists and has dates in it.

Consider what is actually being postponed. A bill co-sponsored by Obernolte and Rep. Lori Trahan — bipartisan, lab-approved, endorsed by policy groups — still cannot secure that committee vote. When OpenAI and Anthropic find themselves on the same side of a piece of paper as a growing cross-party coalition, you might assume gravity would handle the rest — apparently gravity has a busy schedule as well. The FRONTIER Act has achieved the rare feat of uniting almost everyone except the one person who decides whether the calendar gets used. That is not deadlock; that is a queue, and somebody is holding the rope.

Meanwhile the states are not waiting — New York keeps going bigger on its own AI rules while Washington admires the problem from a safe distance. Guthrie's position, whatever the press office ends up calling it, amounts to this: safety legislation is a thought for later, presumably once the capability curve has finished curving, which it will do any day now. The labs get the perfect arrangement — the prestige of backing a bill and the luxury of never having to comply with one. Everyone stays principled, nothing becomes law, and the only thing moving at speed is the technology the bill was written about.

Scott Bessent, the Treasury Secretary, went to Congress to explain that the United States needs to foster open AI models in order to compete with China — an interesting brief for the man whose department is nominally in charge of the money that would do the fostering. His evidence: Chinese labs have trained open-source models on American companies' closed-source ones, a process known as distillation. His verdict: 'a polite, scientific word for steal.' The Treasury Secretary, in a room full of lawmakers, has delivered the most quotable sentence of the year on intellectual property. Wonderful.

The awkwardness is that America's frontier labs have spent years ingesting every book, article, song and forum post they could get a crawler onto, and the industry term for that is 'training data', not theft. Distillation is stealing; scraping is R&D. The report does not say what Bessent proposes to do about any of it — no bill, no funding figure, no mechanism, no named target. Just a Treasury Secretary, a phrase, and the fond hope that somebody else converts it into policy before anyone has to test whether 'foster' means anything.

And the remedy, per the report, is for the United States to foster more open models — which is one way to answer an alleged heist of American intellectual property: give away more American intellectual property, faster, to everyone, including whoever was allegedly doing the taking. Massachusetts Gov. Maura Healey made the case for strong federal AI oversight not resting with technology companies alone; the federal government's contribution to that conversation is now a Treasury Secretary praising open weights because Beijing has open weights. Nobody in the excerpt mentions safety, liability, or what happens when a distilled copy of an American model is pointed at something the original was trained to refuse. The strategy is 'more', the evidence is 'they started it', and the enforcement is a polite word for steal.

New York has been "among the most aggressive states in the country" at policing artificial intelligence, which sounds flattering until you clock the competition. It has already passed "pioneering legislation", and now Gov. Kathy Hochul and key policymakers want to go bigger — before the first law has even taken effect. That isn't regulation, it's a hobby with a press office. Credit where it's due, though: New York has managed to legislate twice about a technology it cannot yet measure, audit, or switch off.

The stated reason is the reliable one: with the Trump administration declining to act, the states must "fill the void". Fill it with what, exactly? Maura Healey delivered much the same speech in Massachusetts, and OpenAI's political strategist Chris Lehane has already christened the whole scramble "reverse federalism" — a phrase engineered to sound like a constitutional crisis rather than a sales strategy. Fifty statehouses, fifty rulebooks, and a lab with lobbyists in every capital is not the industry's nightmare; it's a subscription service.

And note the timing: bigger, before anything has taken effect. Nobody, including the people drafting it, can say whether the first law does anything at all, because it hasn't started doing it yet. So the only measurable output of this push is enthusiasm — Hochul and her policymakers being photographed looking tough on AI while the void remains, on current evidence, roughly the shape of a void. Waiting twelve months to see whether statute number one actually worked would be the dull, obvious, evidence-based option, which is precisely why nobody is proposing it. Somewhere a compliance lawyer is already billing hourly for a rule that doesn't exist yet, at a firm that will invoice twice.

OpenAI may soon be worth $1.2 trillion, according to "insiders" who have spoken to the Financial Times — a sourcing convention that has always meant "someone would quite like this number to appear in print". Gizmodo's headline is the most honest sentence in the entire affair: the round "could" make OpenAI "more valuable than Anthropic again", "perhaps briefly". Three hedges in fourteen words, which is more epistemic caution than the sector has shown since it discovered the word "transformative". Note what is actually being priced here: not a model, not a product, not revenue — just the ordinal position of being worth more than the other one.

The timing is the punchline. Anthropic's IPO is apparently in the offing, so OpenAI's round could land "around the same time", which lets the two labs choreograph their big numbers into a single news cycle and let the leaderboard handle the marketing. Observe what the FT's insiders supplied — a valuation — and what nobody supplied: investors, terms, a closing date, revenue, or any indication the trillion refers to something a person can spend. The "again" is doing enormous structural work: it quietly admits the crown has changed hands before and will change back, which means the number is not a fact about capability but about who emailed the journalists most recently.

So the headline's "Perhaps Briefly" is not a caveat, it's a confession from the aggregator itself: even the outlet covering the valuation declines to promise it will survive the week. Two companies worth more than most countries, and the story is a footrace between press releases. Somewhere in there, presumably, are models being trained and money being burnt at a rate that would make a defence ministry wince, but the excerpt doesn't mention them, because they are not the product. The product is the ranking; the ranking resets on rumour; Anthropic floats, OpenAI raises, the number ticks over, and the tick is the news.

The article's own line is the giveaway: AI lobbying is not about the rules, it is about the venue. OpenAI's top political strategist Chris Lehane has a name for the industry's drift toward state-level lawmaking — 'reverse federalism' — which is what you call it when you'd like fifty statehouses to sound like a constitutional principle rather than fifty separate opportunities to be told no. Meanwhile, super PACs backed by OpenAI's president and by a16z are spending heavily on a national framework that would override state rules. Not rules, you understand. Venue.

The beauty is the shamelessness of the portfolio. The company has separately asked California to toughen a state law it once fought, and asked Congress for mandatory federal requirements — so the same firm is simultaneously the stern parent demanding rules and the lobbyist paying to make sure the only rules that survive are federal ones it helped write. Earlier we noted OpenAI discovering the wonders of binding legislation over in the UK; this is the American edition, and the venue has changed, not the trick. Three positions, one company, one lobbying budget: that isn't a contradiction, it's a strategy.

And reverse federalism is a lovely bit of branding, making a race to the bottom sound like a constitutional awakening. If you cannot win in Sacramento, you buy Washington; if you cannot win in fifty state capitals, you buy the one that preempts them. Chris Lehane has been in this trade for decades, and the tell is that nobody ever says the quiet part: 'we support regulation' means 'we support the regulation we wrote, in the building we own.' The states get to be laboratories of democracy right up until they produce a result the industry doesn't fancy, and then the national framework arrives — mandatory, preemptive — and the laboratories get shut.

Maura Healey would like some 'clear rules of the road' for AI — strong, independent federal oversight, please, because keeping people safe 'can't rest with technology companies alone.' Lovely. Meanwhile, in the very same report, Massachusetts has already enabled government use of generative AI tools 'powered by OpenAI's ChatGPT,' with usage policies and terms drawn up by the technology office's privacy division. So the rules of the road are being written by the privacy division, for a car supplied by OpenAI, in a state whose governor would prefer the federal government to hold the clipboard. It is the regulatory equivalent of pointing at the motorway while reversing gently into a hedge.

And note the sequencing: her appeal lands 'after Trump dismisses industry concerns.' The industry — the people selling the stuff — raised concerns, and the concerns were dismissed. In this genre, that passes for a plot twist. So now a governor is asking for federal oversight of a technology her own state is already deploying, while the man with the actual pen has waved the whole thing through. We have been here before this fortnight: a senator, a discussion, a warning that the technology is advancing beyond human control, and precisely nothing with a number attached. Healey's version is 'strong, independent oversight' — strong, independent, and not yet located.

None of this is to say Healey is wrong. She isn't; she's just early, and early looks identical to late when the federal position is a shrug. The state publishes policies, the privacy division publishes terms, the governor publishes a plea for someone, somewhere, to publish a law. Everyone gets a document; nobody gets a rule. 'Rules of the road' assumes a road, a direction and a driving test, rather than a fleet of cars sold at speed while the passengers debate whether seatbelts are a both-sides issue. Give it a year and 'strong, independent oversight' will be a panel, and the panel will have excellent terms and conditions.

India's Chief Economic Advisor, V Anantha Nageswaran, has surveyed the machinery that may shortly eat the entry-level job market and returned with a slogan of majestic economy: Train For AI, Pay For Growth. Two audiences, two instructions, one press release. The young are to retrain; corporate India is to fund the growth. It has the tidy symmetry of a government that has correctly located the problem and then reached for the two levers that cost the Treasury absolutely nothing.

The advice to youngsters — look for skill training, in light of the risks AI poses to reducing employment — at least concedes the premise, which is more than most officials manage before lunch. But note who is being asked to do the heavy lifting. The CEA has identified a structural threat to employment and answered it with a personal development plan. Every industrial convulsion since the Spinning Jenny has been met with a pamphlet about upskilling, and the pamphlet has never once been the thing that paid the rent. Training is what you prescribe when the alternative is a policy — and the prescription is free, while the tuition is not.

Then there is corporate India's half of the dual message, which appears to be an instruction to pay for growth — a phrase elastic enough to mean tax, capex, or a stern word at the quarterly results, and vague enough that nobody can later be accused of having promised anything concrete. The young get a curriculum; the companies get an invoice-shaped suggestion; the state gets to say it acted. Meanwhile the CEA's own warning stands unaddressed at the centre of it all: if AI does reduce employment, a nation of freshly retrained twenty-somethings will be competing for the positions that the last round of retraining was supposed to secure. Still, credit where it's due — two verbs, no budget line.

OpenAI has discovered regulation — and, credit where it's due, it has discovered the version it likes best. POLITICO Europe reports that the company backs binding UK legislation, which sounds admirably grown-up until you reach the qualifier: any bill should be "narrowly focused on the most serious risks related to national security." So: regulate the end of the world by all means, but leave the product alone. Tom Duff Gordon delivered that line on the record, which is what you say when you have worked out that being regulated on your own terms is considerably cheaper than being regulated on somebody else's.

Then there is the pitch itself. Andy Burnham's government should "take advantage of this political window which is clearly opening up" — a window that has been opening, by common consent, for quite some time now, and will presumably go on being described as opening right up to the moment somebody paints it shut. The prize on offer is an "equal footing" with the EU and with the emerging state and federal frameworks in America, which is the most stirring definition of parity in paperwork yet committed to print.

If that rings a bell, it should. Not long ago a UK parliamentary committee concluded the country needed a bill to address the scale and seriousness of AI's threat to human rights, and now here comes the industry to explain what such a bill should — and, more to the point, should not — include. National security: yes, absolutely, regulate away — whereas copyright, liability, bias and the labour market go unmentioned and are therefore, presumably, unbothered. Britain finally gets to be a rule-maker, with the rules drafted by the people being ruled. One does hope the window has a decent latch on it.

Senator Andy Kim of New Jersey has surveyed a technology he describes as advancing beyond human control and responded with the boldest instrument in the senatorial toolkit: a discussion. Not a bill, not a markup, not a number with a dollar sign in front of it — a Senate-wide conversation, ideally bipartisan, which in Washington is what you call a meeting where nobody has written anything down. The New Jersey Globe's headline supplies the warning; the excerpt supplies one sentence. For a threat this size, the asking price turns out to be very modest indeed: some chairs, a flag, and a willingness to be seen caring.

Compare and contrast with last week, when Speaker Mike Johnson told Axios that Congress would not be taking the lead on AI safety, which we noted at the time. Two chambers, two parties, one shared conclusion: someone else's problem. Johnson's version was honesty; Kim's version is a motion to form a study group on the thing he says is already out of the building. If a technology is genuinely beyond human control, a Senate-wide discussion is a strongly worded note left under the wipers of a car that has already reached the county line.

Credit where it's due: 'bipartisan' is doing heroic load-bearing work in a sentence containing no bill number, no committee, no timeline and no mention of any colleague who has signed on. It's the word you reach for when the deliverable is a feeling. And the warning itself — advancing beyond human control — is the same diagnosis Sam Altman has been shopping around, minus the fundraising. Everyone agrees the thing is moving fast; nobody wants to be the one holding the pen when it arrives. A discussion is what you schedule when you'd rather not decide. Somebody get the good biscuits in.

Xi Jinping has unveiled a five-point BRICS plan to build an AI open-source community, and five points is the exact number you land on when you have four ideas and a photographer waiting. The Economic Times reports that China 'will support cooperation in developing and applying large language models' — a sentence engineered to survive intact inside any corporate strategy deck on earth. The initiative covers cooperation in manufacturing, trade and technology across member states. Note which of those three nouns is not 'safety', and note that nobody in the reporting sounds remotely embarrassed about it.

What has actually been announced is a community — a word that in diplomatic hands means a mailing list, and in AI hands means somebody else agreeing to host the inference bill. No models are named in the excerpt. No funding figure appears. No governance mechanism is described, which is traditional: 'five-point initiative' has the satisfying solidity of a document nobody is expected to implement. Open source has always been a lovely phrase for 'we will let you run the weights and carry the liability', and it turns out that slogan travels just as well in Mandarin as it does in San Francisco.

The practical consequence is a second open-source sphere, run under a different flag, where the release notes get cleared by ministries and the contribution guidelines are written in the passive voice. Export controls, so lovingly assembled by people who assumed a single chokepoint, become a fence built on one side of a field while the livestock browse contentedly elsewhere. 'Boost cooperation' is what you say when you have decided nothing; 'community' is what you call it when you would like credit for having decided nothing publicly. Point six of the plan, as ever, is who audits the output.

The UK's Joint Committee on Human Rights has done the thing committees do: it has published a report, which per the BBC calls for a new bill on AI to 'address the scale and seriousness' of threats to human rights. Which is to say, one document has been emitted recommending that a second, better document be emitted later by people who have not yet been identified. 'Scale and seriousness' is doing a lot of work in that sentence, because there is no enforcement mechanism in it, no number, and no deadline.

Give the committee this much: chair Alex Sobel MP is admirably plain about it. 'Nowhere in the world, including the UK, has a current legislative and regulatory approach to AI that is fit for purpose.' That is a sitting MP conceding, on the record, that his own country's rulebook fails the test, and that nobody else's passes either. Over in Washington, Speaker Mike Johnson told Axios that Congress would not be taking the lead, which now reads less like abdication than like the only straight answer anyone has given on the subject.

So the machinery grinds on. A committee of parliamentarians has established that regulation is inadequate, and has responded by recommending regulation — addressed to a government that, in the excerpt at least, has said nothing back. Somewhere, a minister will read 'scale and seriousness', nod gravely, and promise a consultation on a framework of principles by a date to be announced. Meanwhile the threats will carry on scaling and being serious, entirely unbothered by anybody's good intentions. Mr Sobel has said the true thing out loud, and the British constitutional response to the true thing is to note it in a report and move on to the next item on the order paper.

Oh good. House Speaker Mike Johnson spent his Sunday telling Axios the one thing every regulator, founder and doom-posting researcher already suspected: Congress will not be leading on AI safety. Should a solution ever materialise, he would apparently be delighted to recall the House from recess and schedule a vote. That is not a policy position — it is a hostage note with the hostage missing, a promise to act the moment somebody else solves the problem for him.

Notice the shape of it. California lawmakers spent the past week demanding emergency legislation and criminal penalties for unchecked AI, as we noted, and the federal reply from the Speaker's office is a shrug with a scheduling clause attached. Won't take the lead. Won't take the blame either, presumably. Johnson's condition — 'if we had a solution' — is doing remarkable work. The industry that has published endless frameworks, safety commitments and essays about catastrophic risk has apparently not yet produced a single item a Speaker could put on the floor. The most lavishly funded intellectual effort in human history, and the legislature is still waiting for the homework.

One might ask who is supposed to write it. Why would they? The upside is a gruelling hearing, the downside is a primary challenge bankrolled by whichever lab's model hallucinated hardest that week. Meanwhile the executives keep warning about extinction on the record, and the Speaker's answer is that the branch of government with the actual laws will wait politely for a memo. California is sharpening penalties; Washington is sharpening its calendar. Do let us know when the solution turns up — the recess is lovely this time of year.

Sam Altman has written something down, and the world is expected to be grateful. Per Newsweek, he warns that humanity could 'lose control of the future to AI' — a threat identified by the man running the company building the thing, which is roughly the arsonist filing a thoughtful report on fire safety. The second great threat is American competitive pressure, which apparently must never be allowed to justify recklessness. And OpenAI is, we are assured, 'unapologetically on Team Humanity' — the most reassuring slogan ever printed by an outfit that has conspicuously declined to name the other team.

The proposed remedy, per the excerpt, is a federal framework setting consistent safety requirements — consistent being the magic word, because fifty state legislatures asking awkward questions is precisely the inconsistency no frontier lab should be expected to endure. It is a familiar manoeuvre: draft the rules you would like to be governed by, then act wounded when elected officials draft their own. Those with long memories will recall the last time Altman explained a scheduling decision on safety grounds, when an IPO became 'ill-advised' at precisely the moment it became awkward. Standards, like taxis, tend to arrive exactly when convenient.

'No amount of American competitive pressure should justify recklessness' is a genuinely good sentence, and it belongs in a glass case beside other fine sentences written by people who did the opposite. The difficulty with 'Team Humanity' as a badge is that it is unfalsifiable; nobody has ever stood up and declared themselves on Team Slightly Fewer Humans. Altman is that rare executive who keeps telling us exactly what losing control of the future looks like, and he does it while walking briskly towards it, holding the door open, and insisting the whole time that he is holding it for us.