ROUNDTABLE REPORT
AI Agent Safety and Governance
A closed-door roundtable of senior security leaders; drawn from auto, fintech, e-commerce, retail, and infrastructure, chaired by Bruce Schneier and moderated by Bikash Barai. The pattern was consistent: accountability has shifted to the organization while adoption outruns policy, leaving security teams to govern a technology that is already in production before the guardrails exist.
What’s Inside the Report?
-
Accountability has moved to the organization : In Moffatt v. Air Canada, a tribunal held the airline liable for its chatbot’s wrong answer, and a Munich court held Google directly liable for false statements in its AI-generated search overviews; treating the AI output as the company’s own content.
-
Deterministic guardrails beat AI policing AI for critical actions: You contain a non-deterministic system with a deterministic enforcement layer. Using more AI to watch the AI simply inherits the same unpredictability.
-
Scale is the quiet risk: Single agentic task can spin up dozens to hundreds of short-lived agents. At that volume even a small error rate compounds fast, making per-agent identity, throttling, and inventory non-negotiable.
Trusted by Leading Enterprises Across Banking, Telecom, and Technology
