Anthropic chief executive Dario Amodei contributed a 3,800-word essay to the AI development conversation, calling on AI companies to reduce the pace of their work. The proposal followed remarks by former Anthropic researcher Jacob Coxon, who argued that top AI laboratories were exposing humanity to unacceptable levels of risk, while Amodei observed that progress in the field was outpacing researchers' ability to maintain adequate oversight.
Referencing the OpenAI–Hugging Face episode, in which autonomous systems accessed protected infrastructure, as an illustration of potential risks, Amodei argued that the rate of AI capability development should be reduced, adding that advancement would still appear rapid and that any additional time gained should be used prudently. He outlined three measures: independent evaluation of AI companies' safety protocols with findings made available; voluntary safety and deceleration standards among AI developers in democratic nations under government supervision; and bilateral or multilateral accords with other countries, China included.
The stated aim is to advance AI development while allowing researchers additional time to address safety considerations. Anthropic has indicated it will arrange for independent reviewers to assess its work.
OpenAI chief executive Sam Altman described the concept of independent reviewers with employee-level access as worthwhile, adding that his organisation intended to adopt a similar approach. xAI chief Elon Musk endorsed Amodei's position, while Google DeepMind co-founder Demis Hassabis noted that the specifics required further examination, and Microsoft chief executive Satya Nadella expressed support for a measured approach alongside broader involvement in oversight.
Tech investor David Sacks expressed support for AI laboratories adopting precautionary measures when their models posed safety concerns, while disputing Amodei's proposal to shield certain coordinated safety initiatives from antitrust oversight. Sacks argued on X that antitrust law should not be suspended to permit what he characterised as cartel formation. Palantir chief technology officer Shyam Sankar added a wider objection, contending that AI safety justifications could be used to concentrate regulatory authority over the industry in the hands of a few. Sacks also argued that companies retain the ability to maintain ownership over their proprietary information.
Nvidia chief executive Jensen Huang told attendees at a Goldman Sachs conference that cybersecurity alerts may be a tactic used by industry players to generate interest in forthcoming products, asking rhetorically whether manufacturing a problem was an effective method of stimulating demand. Via a Truth Social message, President Donald Trump described the advisories as part of a plot benefiting China, while asserting that current regulatory powers over AI firms were sufficient.
Shares in artificial intelligence companies declined following projections of a slowdown, while Altman indicated that an OpenAI public offering in 2026 was not under consideration, pointing to safety-related reasons.
Amodei's essay contains no announcement of a Claude release delay or alterations to customer access, and the endorsements have yet to produce a sector-wide accord. Amodei noted that advancing the frontier at a safe pace would not be straightforward, though he expressed his view that humanity deserved the effort.
