Compute Governance, Export Controls, and the Science of AI Safety

Why chips became the chokepoint of AI policy: the US–China export-control arc from October 2022 to the Diffusion Rule and its rescission, compute thresholds as legal triggers, cloud KYC, and data-center politics — then the safety science a governance expert must be able to interrogate: alignment, dangerous-capability evals and their limits, interpretability, and safety cases.

Content current as of 2026-09.

Lessons

  1. Why compute became the lever
  2. The export-control arc: October 2022 to the Diffusion Rule and back
  3. Thresholds, cloud KYC, and the data-center fight
  4. Alignment and dangerous-capability evals — reading the safety science
  5. Interpretability and the safety case