~ / ai-security

AI Security

Attacking and defending LLM apps, agents, and the tooling around them.

How untrusted content in a tool result hijacks an agent's next action, and what stops it.

Poisoning the retrieval store to steer answers and exfiltrate context.