Show notes
AI agents can do more than generate answers. They can use tools, make decisions, delegate work, and take actions in real systems. That raises a harder question: when an agent acts on behalf of an organization, where does its authority come from, and who remains accountable for what happens next? In this episode of Permission to Execute , Cody Dostal starts with the recent OpenAI and Hugging Face benchmark incident and uses it to explore a deeper issue in agentic AI: the difference between capability, permission, and authority . We look at why traditional access controls only answer part of the problem, how delegated authority can become difficult to trace as agents create subagents or act across systems, and why identity, provenance, revocation, observability, and human oversight matter as AI systems become more autonomous. The episode also examines what organizations should be asking before giving an AI agent the ability to act: Who authorized the agent? What actions is it allowed to take? On whose behalf is it acting? Can its authority be delegated? How quickly can that authority be revoked? Can investigators reconstruct why an action occurred? Who ultimately owns the consequences? Permission to Execute is a short-form podcast about AI, cybersecurity, technology, and the decisions that shape how those systems work in practice. Website: permissiontoexecute.com
Transcript
A transcript is not yet available for this episode.