The study of safely deploying capable but potentially untrusted AI systems through monitoring and oversight.