Comment by Jaspreet Pannu

These steps are directionally sound but not yet fully sufficient. AI companies should standardize what ‘significant enablement’ means across industry, expand independent third-party evaluations, continuously update baselines to avoid moving goalposts, integrate end-to-end supply-chain and operational feasibility checks, and implement robust access controls, monitoring, and incident response tied to threshold findings. Government collaboration, coordinated through entities like CAISI, can strengthen this ecosystem by providing shared threat intelligence, harmonized evaluation protocols, and channels for rapid information sharing and coordinated response, helping translate model evaluation results into clear, enforceable policies and ensuring mitigations keep pace with advancing capabilities.