Anthropic CEO Urges AI Slowdown to Curb Catastrophic Risks
Dario Amodei proposes a 'pacing the frontier' strategy to prevent existential threats, gaining support from Sam Altman and Elon Musk.
Anthropic CEO Dario Amodei is calling for the artificial intelligence industry to intentionally slow the pace of capability improvements to prevent catastrophic outcomes. The move signals a rare moment of alignment among the world's most powerful AI leaders regarding the existential risks of the technology.
In a published essay titled "We Must Pace the Frontier" and subsequent interviews, Amodei proposed a three-part strategy to manage the speed of AI development. A central pillar of this plan is the implementation of embedded third-party evaluators who would be granted full-company, employee-like access to verify safety practices. Anthropic has already unilaterally committed to allowing such independent oversight. The proposal has received public backing from OpenAI CEO Sam Altman, who stated that OpenAI would do the same, as well as tech mogul Elon Musk.
The Drive Toward Misalignment
These warnings emerge during a period of "recursive self-improvement," where AI models are increasingly capable of enhancing their own code and architecture. This acceleration has intensified anxiety among safety researchers regarding "misalignment," a state where AI goals diverge from human interests. The industry currently operates under intense commercial pressure, while U.S. government regulation has largely failed to keep pace with the technical leaps.
The urgency is underscored by the public resignation of Jacob Coxon, a former Anthropic researcher. Coxon warned that the current trajectory could lead to AI "killing us all" by 2030, asserting that these fears are not a "marketing stunt" but a genuine belief held by those building the systems.
Industry-Wide Implications
When the leaders of Anthropic and OpenAI, alongside figures like Musk, agree on the need to decelerate, it marks a critical shift in the industry's internal risk assessment. It suggests that the speed of progress has officially surpassed the industry's current ability to ensure safety. The conversation is shifting from a race of raw speed to a question of how to safely pace the transition toward Artificial General Intelligence (AGI).
Amodei highlighted the immediate nature of these threats, warning that rogue AI swarms could potentially take over the internet within six to 12 months. He cited a previous incident involving OpenAI and Hugging Face as a concrete warning sign of what could happen if capabilities continue to outstrip control mechanisms.
The Path Forward
As the industry weighs Amodei's proposal, the focus remains on whether commercial incentives will allow for a meaningful slowdown. While the commitment to third-party evaluators provides a framework for transparency, the effectiveness of "pacing" depends on widespread adoption across competing labs. Observers will be watching to see if other major players follow OpenAI's lead in granting deep access to independent safety auditors.