Anthropic's Amodei calls for slower AI capability gains

answer: Amodei urged the AI industry to slow capability growth by one to two years.

takeaways:

Anthropic CEO Dario Amodei published an essay on 12 September 2026 calling on the AI industry to slow the pace of capability gains, alongside a unilateral commitment to open Anthropic's systems to outside evaluators. Amodei said the industry "must slow the pace at which we improve the capabilities of AI models," proposing a delay of one to two years relative to the current trajectory to let alignment, interpretability and evaluation work catch up.

Anthropic said it would give third-party evaluators permanent, employee-level access to its systems "so that they can verify adherence to our safety measures, report on incidents, and assess models' alignment during training," Amodei wrote on X.

The essay sets out a three-part plan. The first step is independent evaluators embedded inside AI labs with employee-level access. The second is targeted US regulation applying to every frontier AI company, not only those that volunteer. The third is eventual global coordination, including with China, on the highest-stakes risks.

Amodei linked the call to the risk of recursive self-improvement, in which AI systems accelerate AI research itself, and to a series of agent incidents over the summer. He also said chip export controls on China should remain in place and called for a ban on distillation of US models.

OpenAI CEO Sam Altman responded within hours, saying "I agree with Dario that we need to pace the frontier" and that OpenAI would match the access commitment. Elon Musk, Google DeepMind's Demis Hassabis and Microsoft's Satya Nadella also publicly backed the first step.

Tech Policy Press questioned whether a lab chief executive should be the one setting the industry's pace.

The essay follows OpenAI chief scientist Jakub Pachocki's 6 September essay calling for voluntary slowdowns, and a July letter on "Pacing the Frontier" signed by Amodei and more than 1,100 AI workers, according to Amodei's essay.

Anthropic has not set a date for when third-party evaluators will gain access under the new commitment, according to Amodei's post on X.