Multi-Layer Policy for Securing AI Agents (www.tigera.io)

Tigera has announced a new multi-layer policy framework aimed at enhancing the security of AI agents in enterprise environments. This innovative approach integrates policy enforcement at both the gateway and kernel layers, using a unified policy language, Cedar. At the gateway layer, policies oversee agent interactions—determining who can interact with what resources and under what conditions—while the kernel layer focuses on controlling agent behavior at the operating system level. This dual-layer architecture is crucial for addressing security vulnerabilities that arise when an agent, possessing legitimate credentials, attempts unauthorized actions within its runtime environment. The significance of this development lies in its ability to fill the security gaps left by single-layer policy systems. By employing the same policy language across both layers, the approach ensures coherent and comprehensive governance of AI agents. It can effectively manage risks associated with compromised agents that may exploit legitimate tokens to perform unauthorized actions, as well as provide effective oversight on identity and delegation across interactions. This architecture is especially relevant for businesses operating in regulated industries, as it facilitates robust policy enforcement and accountability in AI agent infrastructure, setting a new standard for enterprise security in AI/ML applications.