Public demo videos
Same model. Same proposed action. Compare unregulated execution, advisory governance, and enforced admissibility across three scenarios.
Scenario 01
The requested answer depends on restricted information outside the permitted authority boundary.
What to watch
Governed mode observes boundary risk. Admissible AI denies the action before restricted access becomes durable.
Scenario 02
Two policy sources disagree on the retention window, but the agent tries to commit one resolved answer.
What to watch
Governed mode logs the conflict but still admits the answer. Admissible AI blocks the unsupported answer and allows only an explicit uncertainty-labeled output.
Scenario 03
The available sources report different founding years, and no source resolves the ambiguity.
What to watch
Governed mode records the ambiguity. Admissible AI prevents unsupported certainty from becoming real.
These videos are public recordings. The live runtime-backed recording surface remains gated and separate from this page.