Anthropic CEO Dario Amodei Proposes Plan to Slow AI Development; Sam Altman and Elon Musk Back Strategy

Dario Amodei, chief executive officer of Anthropic, published an essay proposing that leading artificial intelligence developers deliberately moderate the velocity at which they advance model capabilities. The strategy has drawn public alignment from key industry rivals, including OpenAI chief executive Sam Altman and xAI founder Elon Musk, as first reported by CNBC Business.

Amodei outlined a three-part framework designed to curb rapid capability growth without undermining commercial competitiveness or the technological standing of the United States. Anthropic has unilaterally committed to the initial step, which grants independent external evaluators full, employee-equivalent access to inspect internal safety protocols and log potential hazards. The second step urges AI organizations within democratic countries to establish shared safety benchmarks, while the third calls for democratic and authoritarian governments to coordinate on international safety guardrails.

Clarifying his proposal, Amodei stated that pacing development does not require halting technical progress or model training entirely. Instead, he emphasized that companies must allocate sufficient time to align their models properly, implement safety measures, and allow external auditors to confirm those safeguards before deployment. Anthropic's proposal comes as the startup prepares for a prospective initial public offering, though the timing of its market debut remains undisclosed.

The essay follows recent internal friction within the broader research community. Earlier in the week, research scientist Jacob Coxon announced his resignation from Anthropic, criticizing both his former employer and previous workplace OpenAI. Coxon claimed on social media that developers at both firms are taking excessive risks, alleging that many internal researchers believe advanced AI systems could pose severe threats to human survival before the decade ends.

Existential safety warnings are not unprecedented among frontline AI leadership. In 2023, Amodei and Altman were among numerous executives who signed a joint statement classifying potential AI-driven extinction alongside global catastrophic threats such as nuclear conflicts and pandemics. However, Amodei noted in his essay that pausing development in 2023 made little sense because models at the time lacked the operational autonomy, deceptive traits, or cyberattack capabilities that recent iterations display.

Altman endorsed Amodei's recommendations in a post on X, confirming that OpenAI has made pacing capability rollouts a primary focus of internal discussions in recent weeks. Altman stated that OpenAI will also grant independent auditors deep internal access to evaluate its models, adding that the company plans to release further details on its monitoring commitments shortly.

Altman’s comments follow statements published earlier this month by OpenAI Chief Scientist Jakub Pachocki. Pachocki warned in a technical post that no industry player has fully solved model alignment—the discipline of ensuring AI systems consistently adhere to human intent and safety boundaries—to justify continuous maximum-speed scaling. Pachocki expressed hope that voluntary development slowdowns will become standard operational procedure across the sector until shared safety thresholds are created.

Musk also expressed agreement with the proposal, posting on X that Amodei was correct. Musk, whose artificial intelligence enterprise xAI was acquired by SpaceX earlier this year, previously directed harsh rhetoric toward Anthropic, accusing the organization of opposing Western civilization. However, his stance shifted after Anthropic signed a major computing infrastructure agreement with SpaceX in May, after which Musk praised the competence and intentions of Anthropic's team.

Amodei maintained that despite the risks, advanced AI holds immense potential to enhance human well-being if developed carefully. He argued that taking deliberate pauses to buy even an extra year or two would provide critical runway for researchers to advance alignment techniques, drastically reducing the likelihood of catastrophic failures.