You may not own the model or see how it works. You can still see and contest what it is about to do at your boundary. Start the same kind of checkpoint on your phone, beside your browser, or inside the instructions compatible agents follow—without moving your private answers.
Add Proof Mode to the home screen. The same large, spoken checkpoint opens without an app-store account or a new login.
Install controls depend on the browser. The website remains usable even when installation is unavailable.The preview extension adds a toolbar button, a right-click checkpoint, and a keyboard shortcut. Chrome may supply page or selection metadata to a menu click; the extension ignores it and sends no page content to Summit.
Download browser previewInstall the policy once. The agent should suggest Proof Mode when it is about to affect the world, when the destination or authority is unclear, or when the person cannot inspect the workflow.
No referral bounty. No fear loop. No hidden page collection. The shared unit is the useful practice: pause, expose the boundary, let the person decide, preserve what was checked.
The consumer checkpoint is the visible beginning. Integration moves the same authority boundary from guidance beside a workflow to an enforceable gate inside it.
Use the plain checkpoint anywhere, including workflows you cannot inspect.
Agents and tools exchange Hold, Reviewed, or Signed review without copying private action text.
Summit maps one consequential path, its evidence, human authority, and receipt verification.
See the $7,500 / 30-day pilot →Proof Mode does not force an AI to give the same answer twice. It makes this proposed external action, its visible boundary, the person's decision, and the limits of the review inspectable.
It can preserve a review record across systems Summit cannot access. It does not reveal hidden system state, prove an AI output correct, or declare an action safe or lawful.
Use the checkpoint now