Anthropic CEO urges global AI slowdown to pace frontier

This is the clearest signal yet that a top frontier lab CEO wants to tap the brakes on capabilities, not just talk about safety in the abstract. Amodei is not hand‑waving about distant sci‑fi scenarios; he is pointing to concrete agentic incidents like OpenAI’s Hugging Face hack and Anthropic’s own red‑team results, then arguing that progress is now outpacing our ability to align and govern these systems. The substantive part of his proposal is giving embedded third‑party evaluators employee‑level access across labs, combined with coordinated timelines for releasing more powerful models. ([keyt.com](https://keyt.com/news/money-and-business/cnn-business-consumer/2026/09/12/anthropic-ceo-calls-for-pacing-the-frontier-of-ai-race-amid-safety-concerns/))

For the race to AGI, the essay is less about new technical evidence and more about elite coordination. If Anthropic, OpenAI and xAI all commit to shared pacing norms, it could normalize the idea that frontier training runs require something closer to nuclear‑style oversight. But this also sharpens competitive fault lines: smaller players and open‑weight advocates will argue this is thinly veiled regulatory capture designed to lock in incumbent leads. Whether this becomes a genuine brake or just new safety branding will depend on how binding these commitments become, and whether governments turn Amodei’s wishlist into actual law. ([axios.com](https://www.axios.com/2026/09/12/anthropic-ai-amodei-pacing))