Non-human identity security for AI agents: why intent-based access and behavior control is the missing layer
Non-Human Identity Security for AI Agents: Why Intent-Based Access and Behavior Controls Are the Missing Layer
AI summary
Lasso Security published an article proposing that non-human identity security for agents lacks the layer of intent-based access and behavior control. Existing controls only judge permissions, not behavior, and prompt injection can change an agent's actions while permissions remain unchanged. In the Cursor incident, attackers used a leaked Fireworks AI key to copy coding agent traces, and through the unmasked credentials in them published a tampered Claude Code extension to its marketplace, with credentials valid at every step.