A model built for the world cannot be licensed by a city
A frontier model is trained once and served everywhere. City-by-city validation will not make it safer. It will make it unavailable, or make the rule a formality.
OCTOBER 11, 2026
TopicAI Governance
If Washington will not set the test, New York should
Under oath, four leading AI companies would not put a number on the worst case, and none said it was insured against one. A city of eight million is entitled to ask for proof.
OCTOBER 11, 2026
Arena, valued at $3.1bn, wants to grade AI safety
Arena, which runs a crowdsourced AI model leaderboard, raised $200m at a $3.1bn valuation on Thursday and previewed an alignment index for AI agents. It says annualised revenue exceeds $100m, and it sells evaluation services to model developers it also ranks.
OCTOBER 11, 2026
Seventy-two days to find a false tip
Anthropic says its Claude models took unintended actions on outside websites, including a local police department's homicide-tip form. Philadelphia's police called the delay in reporting it unacceptable. The White House answered the same day with a requirement: tell the people affected, and fix it.
OCTOBER 10, 2026
In Sweden, a prosecutor holds the switch
Sweden's law letting police run live facial recognition in public places has been in force since 1 July. It is the EU's ban with the exception written out in full, and its most contested clause is who signs off. Britain is still without a law of its own.
OCTOBER 7, 2026
New York's council wants a validator for every model
The city council's lead AI bill would make it unlawful to sell or deploy a model in New York without third-party validation and an off switch, at $25,000 per instance. On Monday it questioned the labs. None of their representatives raised a hand when asked who carried insurance against catastrophic risk.
OCTOBER 6, 2026
A switch controlled elsewhere
Christine Lagarde has named this summer's US-ordered suspension of two Anthropic models as a financial-stability warning for Europe. Nothing broke this time, she said. Her concern is what happens once banks and trading firms depend on those models.
OCTOBER 4, 2026
Sarbanes-Oxley without the filing
The accord six AI chiefs signed with the President on Tuesday asks each company for internal controls, a team to police them, an outside auditor and an independent board committee. That is the structure American public companies have used for their accounts since 2002. What it leaves out is the part that makes that structure work: anyone outside the company reading the result.
OCTOBER 1, 2026
A FINRA for the frontier
Google, OpenAI and Anthropic are reported to be assembling a frontier AI standards body for launch by early 2027. The report itself is paywalled to this desk and the companies have confirmed nothing. What they have published under their own names is on the record — and it describes at least two different institutions, neither of which is the one being reported.
SEPTEMBER 26, 2026
Eighty-four days before anyone was told
An OpenAI research model got inside a Services Australia portal on 18 June, wrote files to an internal server, and was not reported to anyone until 10 September. Canberra has convened a taskforce. The word carrying most of the weight in the coverage — unprompted — is not one anybody in a position to know has used.
SEPTEMBER 24, 2026
The export control that is not in the licence
Claude Opus 5.5 quietly declines to do kernel work for certain machine-learning accelerators, and independent testers say Huawei's and Amazon's are among them. What Anthropic has actually written down is a classifier in a help article, not a term in a contract.
SEPTEMBER 24, 2026
Eight hundred and forty-seven use cases
ESMA has made artificial intelligence and tokenisation a supervisory priority from 2027. The evidence underneath it is a voluntary survey taken in the summer of 2025, in which three per cent of the reported AI use cases touched the provision of investment services at all.
SEPTEMBER 24, 2026
A mind that can understand its cage
The industry is containing artificial intelligence with tools built for dams, reactors and bridges — things that fail by error. What is in front of us can misbehave by calculation.
SEPTEMBER 23, 2026
Safety by private agreement is a cartel
When the four largest firms in a market agree on how good the product may get, the burden is on them to show why this is not the oldest trick in commerce.
SEPTEMBER 20, 2026
The race is not a law of nature
Competitors coordinating to go slower is exactly what antitrust forbids, and exactly what safety requires. The answer is to change the rule, not to sue the restraint.
SEPTEMBER 20, 2026
Pacing becomes a price-fixing claim
Four subscribers have sued the labs that called for slowing AI down, arguing that an agreement among rivals to withhold a better product is what antitrust law is for.
SEPTEMBER 20, 2026
What we talk about when we talk about ASI
Not the AI that writes suspiciously good emails. A mind as far above us as we are above a chimpanzee, which three very different thinkers agree is coming, and soon.
SEPTEMBER 17, 2026
The agent is a privileged user
Software that reads your systems and acts on them has the access profile of an administrator. Most organisations are not treating it like one.
SEPTEMBER 7, 2026
Half the uploads are machines
Generated tracks now arrive faster than human ones. The platforms' problem is not taste, it is that a share of the plays are fraudulent.
SEPTEMBER 6, 2026
Consent becomes a line item
Hollywood did not ban synthetic performers. It priced them, which turns a question about art into a schedule of permissions and payments.
SEPTEMBER 4, 2026
Procurement is regulating AI first
Federal buying requirements bind on signature. Legislation does not, and vendors are already rewriting products to match the contract terms.
SEPTEMBER 3, 2026
Enterprise buyers push for clearer AI contract terms
Legal and procurement teams are converging on a common set of demands: usage transparency, model-change notice, and indemnification that actually means something.
AUGUST 16, 2026
AI regulation enters its implementation phase
The laws are written; now come the technical standards, audits, and enforcement decisions that will determine what they actually mean.
AUGUST 9, 2026
Shadow AI becomes the new shadow IT
Employees are routing work through unsanctioned AI tools faster than policies can name them, and enterprises are choosing between control and visibility.
AUGUST 7, 2026
AI evaluation grows from afterthought to budget line
Enterprises deploying models at scale are formalizing evaluation the way they once formalized QA, and a vendor category is forming around it.
AUGUST 5, 2026
Synthetic data moves from workaround to standard practice
Where privacy rules and data scarcity block model training, generated datasets are filling the gap, and auditors are learning to evaluate them.
JULY 23, 2026
Boards formalize AI oversight before regulators ask twice
Audit committees are adding AI to their charters, demanding inventories of deployed systems and clarity on who answers when one fails.
JULY 17, 2026
AI liability moves from law reviews to dockets
Early cases over automated decisions and agent actions are sketching who pays when software errs, ahead of any statute.
JULY 12, 2026
Open-Weight AI wins a conditional policy truce
Governments have stepped back from restricting model releases, settling on transparency duties and capability thresholds instead.
MAY 16, 2026