Export Controls and Sanctions as an AI Safety Tool
This page is a structured working draft — real analysis, not yet expanded with the full expert sourcing given to the flagship pages. Safe to build on; treat specifics as provisional until sourced.
The basic mechanism
Export controls restrict the sale of the most advanced AI-training chips and related manufacturing equipment to specific countries, aiming to slow how quickly a restricted state can compound frontier AI capability — the same logic covered in more depth on the Compute Governance page, applied specifically as a geopolitical tool.
What controls have targeted
Since 2022, controls have specifically targeted the most advanced classes of AI training chips and the extreme-ultraviolet lithography equipment needed to manufacture them, rather than broader semiconductor trade — a deliberately narrow target intended to slow frontier-AI-relevant capability specifically.
Documented effects and limits
Controls have measurably slowed but not stopped affected countries’ frontier AI progress; restricted actors have responded with smuggling, indigenous chip development efforts, and algorithmic efficiency gains that reduce how much cutting-edge hardware a given capability level requires. Yoshua Bengio’s own analysis noted that a lower-cost, high-performance competitor model emerging despite controls accelerated the broader competitive race rather than slowing it.
The strategic trade-off
Export controls buy time and slow compounding advantage; they also risk accelerating a restricted state’s incentive to achieve chip independence entirely, which could reduce long-term leverage even as it provides short-term slowdown. Policy researchers are genuinely divided on how this trade-off nets out over a multi-year horizon.
Where this fits in the bigger picture
Export controls are one tool among several covered in this section — most effective paired with, not instead of, the treaty and institutional efforts covered in International Treaties and Summits.