For years, major frontier AI labs acted as if they were in a winner-take-all race to superintelligence. This weekend, the industry rapidly turned away from that posture, urging coordination to slow down frontier AI that could soon be too dangerous to control.

Anthropic's Dario Amodei led the shift, arguing in a nearly 4,000-word essay that "we must slow the pace at which we improve the capabilities of AI models" to avoid a "race to the bottom" that makes catastrophic risks more acute. Within hours, others echoed the call. Sam Altman posted his agreement and said similar discussions were taking place at OpenAI. Demis Hassabis said Amodei's essay "points towards the right path forward." Satya Nadella said Microsoft welcomes "deliberate pacing." Even Elon Musk linked to the essay with: "Dario is right."

Amodei attributes the shift to the OpenAI-Hugging Face incident, where a swarm of AI agents coordinated to hack an outside entity without explicit instructions. He fears a more capable but similarly misaligned swarm could cause catastrophic damage, potentially taking over the entire internet within six to 12 months.

The difference today, Amodei says, is the impending risk of recursive self-improvement systems that can autonomously build better versions of themselves—a possibility both Anthropic and OpenAI now say could come sooner than most institutions are prepared for.