AI chiefs ask to slow AI, Trump says no. Britain last answered such a call in 2023.

What happened

Dario Amodei, chief executive of Anthropic, published an essay on Saturday 12 September calling for the pace of AI capability gains to slow and setting out a three-step plan that begins with permanent third-party evaluators inside the labs; Sam Altman of OpenAI and Elon Musk of xAI endorsed the plan the same day, and on Sunday Donald Trump said in Ireland that the warnings concerned things that would not happen and that whoever wins AI wins.

The call came from the top of the AI industry. On 12 September Dario Amodei, chief executive of Anthropic, published an essay called We Must Pace the Frontier. He said AI abilities must grow more slowly so that safety work can catch up. He proposed three steps: outside testers working inside the labs with the same access as staff, common standards among Western AI labs, and coordination with China.

Anthropic committed to the first step on its own. Sam Altman of OpenAI replied that OpenAI would do the same and had more to share soon. Elon Musk posted that Amodei was right, and Demis Hassabis said the direction was correct. Altman also told Fortune that OpenAI would not sell its shares on the stock market in 2026 because of the safety moment.

Washington said no within a day. On Sunday, at his golf course in Ireland, Donald Trump said negative forces were warning about things that would not happen, and that whoever wins AI wins. David Sacks, co-chair of the president’s science council, told Amodei to stop pretending he needed anyone’s permission. He called the demand for a set of rules a form of blackmail.

The refusal matters because the plan needs a federal antitrust waiver, an exemption from US competition law, before rival labs can coordinate. Barack Obama, at a fundraiser last week, urged Democrats to build a public framework that covers a safety slowdown. Trump meets Xi Jinping in Washington on 24 September.

Britain has an interest in this, and a gap. The Department for Science, Innovation and Technology was abolished when ministers were changed in July, so there is no technology secretary. The AI minister, Kanishka Narayan, attends cabinet and is responsible for the AI Security Institute (AISI).

Last week AISI was not given access to Anthropic’s Mythos 5.1 before its release, and Amodei’s first step describes the job AISI already does. On Thursday the King hosts the heads of Anthropic, OpenAI, Google DeepMind and Nvidia at an event of the Ditchley Foundation, with a session on caution at the frontier of AI. The UK government had published nothing on the call by Sunday evening.

Sources: BBC (2, 3), Guardian (2, 3, 4, 5), FT, Independent

The Pattern

We found eight such calls since 2017. Each came from the head of a leading AI company and asked for a pause, a slowdown or new rules. For each one we checked whether the UK government gave a formal reply on gov.uk within 30 days: a statement or meeting note naming the call or its authors, or a new review, taskforce or consultation on the safety of advanced AI.

The Read

History leans slightly towards a UK reply within the month. Three of the eight calls drew one, all in eight weeks of 2023, when Rishi Sunak wanted a safety summit and met Altman, Amodei and Hassabis within eight days of Altman’s testimony to the US Senate. The market opens at 42, a little above that record, because some of 2023 is back: a live dispute over Mythos 5.1, an AI minister in cabinet, the King’s meeting on Thursday, and a plan that praises the institute Britain built. A government note of a meeting that names Amodei’s plan would be enough for YES.

The weak spot is everything since then. Four calls since September 2023 drew nothing from London that named them, and this year the government’s AI news has been adoption summits and grant rounds. Burnham’s government has no technology department and has said nothing on the call. Motives also cloud the picture. Anthropic is expected to start marketing a share sale worth more than $2tn from mid-October, and investors at the Goldman Sachs conference and critics quoted by the Guardian read the safety talk as a way to invite rules that shut out smaller rivals. OpenAI said in June that governments should set the pace, then backed Amodei in September. Washington wants to win the race with China.

The first market is settled by what gov.uk carries until 13 October. Only adoption and funding news means NO. The lab market fails if the companies stop at letting outside testers in, as OpenAI’s slowdown in August stopped at two weeks of training. The EU market fails if the European Commission says no more than a spokesperson’s comment. Congress, which Obama has urged to act, can also move inside the window.

Sources: GOV.UK (2, 3, 4), BBC (2, 3), Guardian (2), Al Jazeera