Cybersecurity Experts Warn Against Overprovisioned AI Agent Access

Cybersecurity experts at the RSAC Conference 2026 warned that the proliferation of autonomous AI agents is creating critical security vulnerabilities through non-human identities and overprovisioned access. Rama Sekhar, a partner at Menlo Ventures, stated that AI agents are shifting enterprise security from a tool-based model to an autonomous operating layer. He argued that manual security approaches are unable to keep pace with AI-driven attacks, urging a transition from simple detection to automated remediation and just-in-time access to limit the potential blast radius of agentic AI.

Similarly, Danny Brickman, CEO and co-founder of Oasis Security, asserted that traditional identity and access management systems designed for humans cannot support the rapid growth of machine identities. Brickman noted that organizations frequently grant excessive permissions to AI agents, leaving security officers with limited visibility. To balance productivity with risk mitigation, he advocated for a shift from role-based to intent-based access control, supported by session-based access and guardrails.