Solutions
One primitive. Five places it changes the answer.
Every one of these starts from the same gap: between the last identity check and the executed action, your stack is making an assumption. Each page below is where that assumption costs something specific, and what replaces it.
Action Authorization
A session can be delegated to an agent. Authorization cannot.
An agent inside a valid session can obtain every credential it needs to act. It cannot obtain a fresh biometric press from an enrolled human on a bound device. That press is the gate.
Teams putting agents on destructive, financial or data-egress operations.
The agent mandate
The agent gets autonomy. The human keeps the boundary, signed.
Signed, scoped and expiring — the agent operates inside a bound a person sealed.
Autonomous workflows that cannot pause for a human on every call.
Protected screens
A signature over a request the user never saw proves nothing.
The approval surface is rendered outside the calling application's control, and what the person saw is what their device signs. The approval carries that content rather than a reference to it.
Payment confirmation, clinical sign-off, any approval where the display is the control.
Privileged access
The session gets them in. The operation waits for the person.
Elevation stops being a state that lasts and becomes a fact about one operation. The consequential ones wait until they carry an approval the enrolled operator released on their own phone.
Production access, secrets, break-glass, permission changes.
Payment authorization
Authenticating the payer is not authorizing the payment.
A signature computed over the payee and the amount, plus a hash of what the payer was shown. Dynamic linking that is cryptographic rather than procedural.
PSD2 Article 97, APP scam liability, agent-initiated purchases.
See it on your own flow.
Your app, your call, our SDK.