UK-Germany Joint Statement on advanced AI safety and security
policyAISI blogJul 2, 2026
Standard evaluations cap how much compute AI agents can use. We show that raising those caps changes measured capability, the difficulty of tasks agents can solve, and how fast the frontier appears to move.
Impact: MediumConfidence: LowNovelty: Material
policysource-feed-c8d6a7cd-73db-479f-87c4-d0961ba02191agentssafety