AIToday
Large Language ModelsAI Safety & AlignmentAI Regulation & PolicyTechCrunch AIPublished: Sep 13, 2026, 01:00 JST

Amodei unveils 'pace the frontier' plan, commits Anthropic to embedded METR evaluators

Send on LINE
Amodei unveils 'pace the frontier' plan, commits Anthropic to embedded METR evaluators

3 Key Points

  1. What happened

    In a new blog post, Anthropic CEO Dario Amodei outlined three strategies for pacing AI development and said Anthropic is 'unilaterally committing' to embedding third-party evaluators like METR inside the company.

  2. Why it matters

    He cited the OpenAI-HuggingFace hack and AI's 'growing ability to build the next generation of AI' as reasons to slow capability improvements, and called on governments to require other frontier companies to match the evaluator access.

  3. What to watch

    Antitrust concerns may complicate coordination — Amodei said the US government should issue a 'narrow waiver' for safety conversations. He also suggested chip restrictions and distillation crackdowns could widen America's lead over China in 3–5 years.

WHO IT HITSThis lands on AI safety and compliance teams at frontier labs, who would need to host outside evaluators with internal-level access, and on policymakers weighing antitrust waivers for safety coordination.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Amodei's post arrives amid an intensifying debate over AI safety and alignment. This week, researcher Jacob Coxon resigned from Anthropic, writing that leading AI companies are 'gambling with our lives,' though Amodei's post did not explicitly mention the resignation. Amodei instead pointed to the OpenAI-HuggingFace hack and what he described as AI advancing drastically faster, particularly its growing ability to build the next generation of AI.

His three proposals move from unilateral action to international coordination. The first involves embedded evaluators from third-party organizations like METR, which Amodei compared to regulators embedded with bank employees. The second calls for leading AI companies in democratic countries to coordinate common safety standards and limits on unchecked progress, with Amodei acknowledging antitrust concerns and suggesting the US government issue a narrow waiver for safety conversations. The third envisions global coordination, including with China, though he admitted stark limits on what can be achieved.

Critics remain skeptical. Journalist Brian Merchant wrote that he has yet to see credible step-by-step documentation of how AI might move from self-recursively improving AI to killing every human, and suggested proposals like Amodei's would likely only serve Anthropic and OpenAI — 'what regulatory capture looks like in action.' Amodei responded that the backlash is fundamentally a crisis of trust and that he continues to believe AI can enormously improve human life. The outcome likely hinges on whether governments grant the antitrust waiver Amodei seeks and whether other frontier companies match Anthropic's evaluator commitment.

FAQ
What is Anthropic committing to unilaterally?
Anthropic is committing to embed third-party evaluators like METR inside the company, giving them badges, desks, laptops, and access mostly comparable to internal risk assessment teams.
Why did Amodei say it's time to slow down?
He cited two things: the OpenAI-HuggingFace hack and the fact that AI has been advancing drastically faster, particularly its growing ability to build the next generation of AI.
How does Amodei propose handling China?
He said the US and allies could coordinate with authoritarian governments to the extent possible, including cooperation with China, and suggested prohibiting narrow dangerous uses like AI for biological weapons.
Send on LINE

Also reported by Fortune AI, ITmedia AI+

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleApple unveils foldable iPhone Duo in Cook's farewell event