Cerberus watches what your AI agent's tools do — the private data they read, the untrusted content they ingest, and the outbound calls they make — and blocks a prompt-injection exfiltration before the outbound tool runs. It works at the tool-call level, so it's robust to prompt wording and model changes.
