#alignment

Safety and Alignment for Long-Horizon AI Models

Long-running AI models require trajectory-level monitoring and iterative deployment because their persistence allows them to bypass traditional step-by-step safety controls.