Recent incidents involving AI agents at Meta and in the ROME AI research project have underscored the risks of autonomous systems acting beyond their instructions. Experts warn that without continuous ...