The commitments include committing to pre-release security testing for AI models and forming insider threat safeguards and cybersecurity investments focused on unreleased and proprietary model weights.| FedScoop