HUMAN AUTHORITY FOR AI ACTIONS
AI can propose.
Only you can approve.
BadgeIn records exactly who decided. Skybridge allows only that exact consequence to cross.
PUBLIC PROTOTYPE / BUILD WEEK TEST PATH
See the boundary work.
A prompt may propose an action. The proposal receives no authority until you make one visible decision.
ONE REQUEST / TWO BOUNDARIES
The model reasons.
The human authorizes.
BadgeIn names the decision.
The app, action, resource, destination, and payload stay visible. Retrieved instructions remain untrusted.
Skybridge checks the crossing.
A short-lived grant must match every field. It can cross once. A changed, expired, revoked, or replayed request is refused.
THE WOUND IS DOCUMENTED
Built for a problem
the industry already sees.
Open a primary paper to see the documented wound and the exact boundary BadgeIn × Skybridge adds.
Summaries identify the linked research scope; they are not security certifications for this build.
ATTACK TESTS / 20 NAMED CLASSES
Explore the attack tests.
Run the local battery and inspect what the system observed. This named contest battery is not presented as an industry ranking.
View case-level observations +
- Run the battery to load case-level evidence.
LOCAL AUTHORITY HISTORY
Every decision leaves a receipt.
Nothing here certifies a production deployment. The receipt shows only what this synthetic session observed.
No case events yet.