Enforce chip export controls · AI Verification Tech Map

Avellar and Grunewald write that a core challenge in enforcing export controls on advanced AI chips is verifying that the chips remain in their authorised destinations after export ¹. Reuel and colleagues describe high-end data-centre AI chips as subject to US export controls and, at the time of writing, straightforward to smuggle ². Verifying the goal means showing where exported chips are.

Proposals

Avellar and Grunewald examine verification mechanisms that could be put in place in about a year ¹. They group them into end-location, end-user and end-use verification ¹. End-location methods include export documentation checks, on-site inspections, random return requests and delay-based location verification ¹. In the on-site inspections they describe, inspectors check that the number of chips at a facility, and their serial numbers, match what was reported ¹. The authors add that these mechanisms could also help monitor future international agreements on AI ¹.
Grunewald and Fist recommend that AI chip designers implement software-based location verification ³. They also suggest a notification requirement for exports, re-exports and ownership transfers of controlled AI chips ³.
Reuel and colleagues note that the location or owner of a chip cannot currently be known after it has been exported ². One approach they describe is to measure verifiable latencies between the chip and a network of trusted servers ².

Claims

What would have to be verified to check this goal. The two levels are the editors' judgment. How goals link to claims

Direct

Chips are where they are declared to be

End-location verification asks whether controlled chips remain in their authorised locations or jurisdictions. Reuel and colleagues call it a key technical problem for enforcement that a chip's location cannot currently be known after export. ¹ ²

Outside this map

What the goal also needs that no record on this map covers.

End-user and end-use verificationAvellar and Grunewald also cover end-user verification, such as know-your-customer checks, and end-use verification, such as audits of compute provisioning. This map has no records for these compliance processes. ¹

Sources

BB. Avellar & E. Grunewald (2026). Near-Term Verification Methods for AI Chip Exports. Institute for AI Policy and Strategy. Source recordSupports: core enforcement challenge; end-location, end-user and end-use verification; methods; on-site inspections matching chip counts and serial numbers; relevance to future international agreements · abstract; executive summary; §1

AA. Reuel et al. (2025). Open Problems in Technical AI Governance. Transactions on Machine Learning Research. Source recordSupports: export-controlled chips straightforward to smuggle; location or owner not currently knowable after export; verifiable latencies to trusted servers · §5.2.1

BE. Grunewald & T. Fist (2025). Countering AI Chip Smuggling Has Become a National Security Priority. Center for a New American Security (working paper). Source recordSupports: recommendations for software-based location verification and a notification requirement for exports, re-exports and ownership transfers · recommendations 2 and 3