3 Comments
User's avatar
Pawel Jozefiak's avatar

'60% of organizations lack formal agentic AI risk assessments' while agents are already running in production - the governance is lagging the deployment by months. The 90-day path (inventory → detection → governance) is realistic and correctly ordered. What I'd add before day 1: map which human credential scope each agent inherits in your environment. That single-person-equivalent access scope becoming an attacker's access scope is the risk most security red teams miss when evaluating AI deployments. The credential mapping step alone changes the risk conversation: https://thoughts.jock.pl/p/building-ai-agent-night-shifts-ep1

Fernando Lucktemberg's avatar

Hey Pawel, thanks for the comment, and completely agree that access scope of agents are being overlooked. That's something I'm expanding on probably next Thursday, I'm finishing a piece called "Agent-Native Identity: How AI Agents Authenticate, Authorize, and Trust Each Other". Would love to hear your opinion on that one. :)

Pawel Jozefiak's avatar

Can’t wait for that piece!