Agent authorization gap: why verified agents are still a risk

Cisco’s security chief says verified AI agents can still become a major risk if they’re allowed too much access after login.

At RSAC 2026, Anthony Grieco said customers are seeing rogue agents that pass identity checks but then read unauthorized data or take unapproved actions.

The core gap is authorization, not authentication: knowing an agent is real does not mean it should be trusted to do everything it requests.

Standards groups and vendors, including NIST NCCoE, OWASP, the Cloud Security Alliance, Duo IAM, and MCP gateway efforts, are working on the problem, but no solution is complete yet.