How to Force a Slowdown in AI Model Development

WIRED explores how a forced slowdown in frontier AI model development could work in practice, drawing on Raymond Douglas's report 'Pacing the Frontier' on external model audits, data center monitoring, and agreements with China.

Compiled by Sergey KostenchukPublished 2026-09-20Updated 2026-09-20

2026-09-20 Safety OpenAI

🛡 How to Force a Slowdown in AI Model Development

WIRED explores what forced oversight of the development of such models would look like. The foundation is Raymond Douglas's report 'Pacing the Frontier': the heads of all major labs support some form of pause, but oversight mechanisms are not yet in place.

🌍 External model audits, data center monitoring based on billing and GPU utilization, 'kill switches' in chips, and agreements with China are being discussed. For clouds and vendors, these are future compliance requirements for large-scale training. Acuteness: Claude already performs 26% of Anthropic's AI research (0% at the beginning of 2026).

👤 If the mechanism is implemented, the pace of new model releases and the rules for using cloud GPUs will change. The report at pacing.tech provides a map of options — from training reporting to chip shutdown.

Source 1: https://www.wired.com/story/heres-how-an-ai-slowdown-could-actually-work/

Source 2: http://pacing.tech/

Sources