Amodei called for a slowdown and Altman shelved OpenAI's IPO
Dario Amodei published an essay on Saturday asking the industry to pace capability gains, and committed Anthropic to giving third-party evaluators permanent employee-level access to its systems; the same day Sam Altman said OpenAI would match that commitment and ruled out a listing this year, calling an IPO now ill-advised.
Amodei called on the AI industry to slow down Anthropic chief executive Dario Amodei published an essay, “We Must Pace the Frontier”, on Saturday, arguing that the rate of capability gain has to be slowed so that risk prevention can keep up, with a three-part plan covering independent evaluators, shared safety standards and limited international coordination. Anthropic is unilaterally committing to the first of those steps: permanent, employee-level access to its systems for third-party evaluators, so they can verify its safety measures, report on incidents and assess how its models are aligned during training. Source
Altman and Musk backed the slowdown within a day Sam Altman posted the same day that he agrees the industry needs to pace the frontier, that independent evaluators with employee-like access are “a great idea, and we will do the same”. Elon Musk’s response to the same essay was “Dario is right”. Source
OpenAI will not float this year, Altman said In an interview with Fortune published on Saturday, the OpenAI chief executive said that given everything happening on the safety front this would be an “ill-advised moment” to list, pointed to 2027 as the more realistic horizon, and said the company feels no pressure to move fast. He also said OpenAI and the other leading labs may be close to announcing a pact to slow development and address safety risks together. Source
Anthropic told investors it will be profitable again The Financial Times reported on Sunday that Anthropic has told shareholders its adjusted operating income will be positive for a second consecutive quarter, on gross margins above 80 per cent before revenue shared with distribution partners including Amazon and the cost of training its models. It has picked Nasdaq for the listing; second-quarter revenue was $11.5bn, and annualised revenue reached $65bn at the end of July. Source
Nadella backed pacing and put Microsoft’s model rules out for comment Microsoft chief executive Satya Nadella posted on Sunday that the company welcomes the “deliberate pacing” needed to make alignment the design goal and likes the idea of “embedded evaluators”, and said Microsoft would publish the code of conduct underlying its own MAI models the following day for public consultation. Source
Speaker Johnson said Congress will not lead on AI safety Mike Johnson told CNN’s State of the Union on Sunday that “if Congress just races in and does some sort of emergency session to try to regulate AI, we will lose the race to China”, and that he would rather the companies took responsibility for the safety of their own products. He said he wants to convene a meeting between President Trump, members of Congress and the AI chief executives instead. Source
A DeepMind safety researcher left for METR Josh Engels said on Sunday that he left Google DeepMind’s AGI safety team three weeks ago for the independent evaluation group METR, turning down offers from Anthropic and OpenAI, because he sees a “terrifying chance” that AI systems cause immense harm within five years. He is the second frontier-lab safety researcher to go public with that reasoning in a week, and METR is the same organisation Anthropic hired on 9 September to investigate four incidents involving its own models. Source
Grok 4.7 needs “a few more days”, Musk said Elon Musk said on Friday that the model needs “a few more days” of reinforcement-learning work, because the training run over-penalised response length, so it gives up too early on hard problems it can actually solve, and it is not rigorous enough about checking its own answers. Grok 4.6 remains xAI’s newest released model. Source