AI Safety Week | Day 3: Verifying Frontier AI Development: Canada’s Role in Building AI Verification Technologies

International agreements run into a trust deficit. A country, or a company, will only accept limits on its own AI development if it can be confident that others are keeping to the same limits. Arms control faced the same problem and solved it with verification: inspections, monitoring and technical measures that let each side check that the other is complying without having to trust it. For AI, most of these tools don’t exist yet. We would need ways to track where advanced chips go and how they are used, to see what is running inside large data centres, and to confirm what a model was trained to do, all without exposing commercial or national-security secrets.

Canada is well placed to help build these tools. It has strong research in cryptography and AI security, an established satellite and earth-observation industry, and decades of experience with nuclear safeguards and arms-control diplomacy. This working session of about 20 people from government, the security community, research, funders and industry will identify which verification technologies Canada could lead on, who would build them and where the additional talent would come from. It opens with a 15-minute expert briefing and questions, followed by breakout work facilitated by international leading experts in verification technology and a round in which the group prioritizes ideas across different verification areas.