AI safety is the discipline of identifying and reducing risks from AI systems, including reliability failures, misuse, harmful outputs, loss of control, and unintended consequences.
Safety practices include testing, red teaming, monitoring, access control, fallback procedures, and governance before and after deployment.