In this episode of the Copilot Show, the host interviews Mehrnoosh Sameki from Microsoft about the evolution of AI, focusing on the shift towards agentic AI and its implications for responsible AI practices. Mehrnoosh discusses the challenges of ensuring safety and alignment in AI agents, including task misalignment, prohibited actions, sensitive data leakage, and agent hijacking. She introduces Microsoft's Foundry Control Plane as a solution for managing and governing AI agents, emphasizing the importance of guardrails, observability, and unique identities for agents. The conversation also covers the role of red teaming in identifying AI vulnerabilities, the augmentation of human skills with AI, and the importance of regulations in the AI space, highlighting Microsoft's efforts to translate regulations into actionable requirements for AI development.
Sign in to continue reading, translating and more.
Continue