Definition
A hardened execution environment that runs an AI agent with a verifiable identity and a record of everything it did, rather than only trying to isolate it. It gives an agent a trustworthy identity to act under and an audit trail, fixing the delegation and accountability gaps that blocked mainstream adoption. It moves agent security from sandboxing to identity-and-intent containment.
Why it matters
You can't hold an agent accountable or prove what it claimed to do if you don't control the runtime it executes in.