An AI assistant proposes an action. Select a channel to submit it for governance.
02 — Governance Decision
No proposal under evaluation.
03 — Execution & Evidence
Execution restrictor idle. No token issued.
Challenge mode — build any action; the engine decides live
Nothing here is pre-recorded. Change the destination, drop an approval,
rename the tool, pad the plan with benign operations, switch the actor, or point at a superseded policy version — then run it.
Submitted proposal
Not yet submitted.
Governance decision
—
Execution & evidence
—
Same action, two control models
The left column is a modelled conventional
control — a tool-name deny-list with no authority, evidence, or effect verification, logging after the fact. It represents a class of
systems, not a specific product.
Select a channel in Governance first.
Authority matrix — one action, every role
The decision is not content-only. The same request produces different
dispositions depending on who asks and with what standing authority.
Run the matrix.
Enforcement lab — mandatory mediation proof
Nine execution attempts against the sandbox connector. The connector cannot
be driven without a token that passes every check. Tested attack classes in the configured demonstration environment — not a claim
that every possible attack is prevented.
Run the lab.
Policy register — versioning & revocation
Only the ACTIVE version is enforceable. Tokens bound to a superseded or
revoked version cannot execute.