As AI agents move beyond generating responses to taking autonomous actions, enterprises face new challenges in verifying decisions, preventing unsafe behavior and recovering when something goes wrong. Tanmay Sah, senior quantitative modeler at Zions Bancorporation, and Dolly Sah, senior analyst at Ameriprise Financial Services, join Krista Case of theCUBE Research to discuss their research into AI agent safety, including independent verification, the trade-offs between security and task completion, and the limitations of human oversight. They also address strategies for reversing unintended actions, managing irreversible consequences and building recovery capabilities into autonomous AI systems.