Threat Model

Threat Model

The primary adversary is untrusted content reaching an agent that can act: prompt injection in web pages, emails, messages and files, plus stale or tampered state feeding an action path.

Current proof coverage is bounded to the tested threat model. Residual risk is documented.