Anthropic CEO Amodei calls for AI speed limits and embedded company auditors
Anthropic CEO Dario Amodei published a roughly 3,800-word essay calling for a controlled slowdown in AI development, warning that recursive self-improvement could threaten the entire internet within six to twelve months if capabilities outpace human control. He proposes permanently embedding independent auditors inside AI companies with access to internal systems and rights to publish findings, plus shared safety standards and global agreements modeled after SALT disarmament treaties. Anthropic says it is unilaterally committing to grant third-party evaluators permanent, employee-like access to verify safety adherence. Major outlets including the New York Times, BBC, and Le Monde covered the call. The six-to-twelve-month timeline remains speculative, and earlier reporting on massive compute commitments complicates how slowing rhetoric aligns with industry scaling.
Anthropic CEO Amodei calls for AI speed limits and embedded company auditors
Anthropic CEO Dario Amodei is calling for a controlled slowdown in AI development. Amodei proposes that independent auditors should be permanently embedded inside AI companies, with access to internal systems and the right to publish their findings.
Key takeaway
Amodei's essay and unilateral third-party access pledge move safety governance from debate toward embeddable oversight at a frontier lab.
What happened
In a 3,800-word essay reported by the New York Times, Anthropic CEO Dario Amodei called for greater safety controls across the AI industry and a controlled slowdown in development. The Decoder reports he warns recursive self-improvement could threaten the entire internet within six to twelve months and proposes embedded auditors, shared safety standards, and global agreements modeled after SALT disarmament treaties.
Techmeme cites Dario Amodei saying Anthropic is unilaterally committing to give third-party evaluators permanent, employee-like access to verify adherence to safety measures. Coverage from BBC, Le Monde, and other outlets frames the statement as a public call to slow AI development, while The Decoder separately notes Anthropic reportedly signed compute contracts worth up to $517 billion in eleven months.
Evidence
Amodei published a 3,800-word essay calling for greater industry safety controls.
NYTimes Technology · attributed
In a 3,800-word essay, Dario Amodei laid out the rapidly advancing capabilities of artificial intelligence and said there needed to be greater safety controls across the industry.
Amodei warns recursive self-improvement could threaten the entire internet within six to twelve months.
The Decoder · attributed
He warns that recursive self-improvement could threaten the entire internet within six to twelve months and proposes embedded auditors at AI companies, shared safety standards, and global agreements modeled after the SALT disarmament treaties.
Anthropic is unilaterally committing to permanent third-party evaluator access.
Techmeme · attributed
Amodei says Anthropic is "unilaterally committing" to giving third-party evaluators permanent, employee-like access to verify its adherence to safety measures
Amodei proposes independent auditors embedded inside AI companies with publishing rights.
The Decoder · attributed
Amodei proposes that independent auditors should be permanently embedded inside AI companies, with access to internal systems and the right to publish their findings.
Anthropic reportedly signed compute contracts worth up to $517 billion in eleven months.
The Decoder · attributed
Anthropic signed compute contracts worth up to $517 billion in eleven months but still trails OpenAI's $750 billion plan through 2030.
Why it matters
Frontier labs are competing to define safety standards before governments act, which could reshape compliance costs, audit requirements, and release timelines across the ecosystem.
Limits and uncertainties
The six-to-twelve-month recursive self-improvement timeline is speculative and lacks public technical evidence in the reporting.
The NYTimes piece is an opinion essay, not a formal policy proposal with enforceable industry rules.
Techmeme reporting does not specify which third-party evaluators will receive access or how conflicts of interest will be managed.
Practical implications
Operators should track whether slowing rhetoric translates into delayed model releases or stricter internal safety gates at Anthropic and peer labs.
Builders may face rising compliance expectations if embedded auditors and third-party access become de facto enterprise or regulatory prerequisites.
What to watch
Whether Anthropic names third-party evaluators and publishes audit findings under its unilateral access commitment.
If regulators or industry groups adopt embedded-auditor or SALT-style treaty proposals from Amodei's essay.
How Anthropic's reported $517 billion compute commitments align with its public call to slow development.
