The latest Frontier Safety Framework addresses AI risks including harmful manipulation and misalignment, aiming to protect operators and high-stakes contexts.