Capture Notes

Paper on fragility of denylist-based controls in AI agents.

AI security relevance:

• Supports research gaps around brittle policy controls and command equivalence in tool-using agents.
• Useful for agent least-privilege and runtime policy topics.