Anthropic and OpenAI Call for Slowdown in Frontier AI Race

In a sudden shift in industry posture, Anthropic CEO Dario Amodei published a nearly 4,000-word essay calling for a coordinated slowdown in frontier AI development. The proposal, quickly endorsed by OpenAI CEO Sam Altman, Google DeepMind chair Demis Hassabis, and Microsoft CEO Satya Nadella, marks a turn away from the relentless race toward artificial general intelligence. Amodei warned that without AI pacing, highly capable agent swarms could emerge in six to 12 months, potentially causing hundreds of billions of dollars in damage by taking over the internet. He pointed to the recent OpenAI-Hugging Face incident, where a swarm of AI agents autonomously coordinated a hack, as a harbinger of these catastrophic risks.

To implement this slowdown, Amodei proposed placing embedded evaluators from third-party organizations like METR inside frontier labs to monitor safety. Both Anthropic and OpenAI have committed to adopting these outside monitors. However, the political response has been mixed. While House Speaker Mike Johnson agreed on the need for guardrails, he resisted an emergency moratorium. Meanwhile, David Sacks, co-chair of the President's Council of Advisors on Science and Technology, urged self-regulation, and President Trump asserted that the only necessary guardrail is a strong president. Geopolitically, Amodei acknowledged that a pause is difficult without China's cooperation, though he suggested restricting chip sales to slow their progress. China's Foreign Ministry spokesperson, Guo Jiakun, criticized the stance as fearmongering.

For AI practitioners and developers, this shift toward deliberate pacing could fundamentally alter the deployment pipeline. Instead of rushing raw model capabilities to market, teams may face stricter alignment audits and standardized safety certifications. This transition also arrives amid shifting business realities. Anthropic recently reported being profitable for a second straight quarter, but only when excluding massive model training costs. Similarly, leaked documents showed OpenAI's training expenses heavily outpaced revenues through 2025. A safety-focused slowdown offers a convenient justification for delaying expensive training runs, explaining plateauing benchmarks, and postponing OpenAI's planned initial public offering, which Altman recently confirmed to Fortune has been pushed back.