AI Frontiers proposes human inspectors for verifiable US-China AI slowdown
Human inspectors, not just AI, could verify an international AI slowdown—here's how.
In the wake of over 1,000 AI employees urging the US government to support an international effort to slow frontier AI development, OpenAI and Anthropic have endorsed the call. But skeptics argue a slowdown is unworkable because verifying compliance—especially against China—would require privacy-preserving monitoring tech that doesn't yet exist and could be bypassed by nation-state attacks like laser bit-flipping or key extraction. AI Frontiers argues this misses the point: robust verification is already possible with low-tech methods.
Their proposal, Whole-Lab Inspection (WLI), would place human inspectors from both the US and China inside frontier AI companies. Inspectors would receive broad physical access plus read-only access to communications, code repositories, and compute telemetry—essentially the visibility of a CISO without operational power. Using monitoring systems companies already run, WLI could detect unsanctioned training runs or suspicious activity, making a joint slowdown enforceable today. The main barrier isn't technology, but political will.
- WLI (Whole-Lab Inspection) uses human inspectors with physical access and read-only telemetry to verify AI development limits.
- Over 1,000 AI employees and both OpenAI and Anthropic have endorsed an international effort to pace frontier AI development.
- Skeptics' concerns about privacy-preserving verification are dismissed—political will, not tech, is the real bottleneck.
Why It Matters
If adopted, human-inspector verification could enable enforceable AI slowdown agreements between the US and China, changing global AI governance.