TrustLogix CEO Proposes Runtime Authorization to Secure AI Agents

Ganesh Kirti, Founder and CEO of TrustLogix, is calling for a fundamental shift in AI security, moving away from identity-based permissions toward runtime action authorization. Kirti argues that traditional three-tier application models cannot secure AI agents, which frequently operate under shared service accounts and remain vulnerable to prompt injection attacks.

This proposal comes as Gartner predicts that 25% of enterprise security breaches will result from AI agent abuse by 2028. To mitigate this risk, Kirti proposes a framework utilizing least-privilege policies derived from specific agent missions and data classifications. He recommends that organizations extract intent from immutable business events rather than agent prompts and centralize decision authority to prevent policy conflicts.

As part of this security architecture, Kirti suggests implementing granular kill switches capable of revoking grants across all connected systems. He advises organizations to start by applying monitor-only policies to a single workflow involving regulated data to evaluate revocation speed and decision latency. This effort aligns with broader industry standardization, such as the OpenID Foundation's finalization of the Authorization API 1.0 in January, which standardizes how enforcement points request security decisions.