Try live
Bind an agent's stated intent to its concrete output inside a carrier image — key-bound, fail-closed on wrong key. Not proof of model honesty, not undetectable stego.
Experimental audit-grade binding — not proof of model honesty, not lawful-process override, not undetectable steganography.
Demo host: isolated demo host set by your deployment
Intent-Output Binder · Agents & policy
Bind intent to output carrier
Bind an agent's stated intent to its concrete output inside a carrier image, then extract and verify the binding with key-bound fail-closed semantics. The demo host runs in isolation — nothing is stored.
Not proof of model honesty, not undetectable steganography. Sample data for demonstration only.
Carrier image
01No carrier loaded. Upload an image or generate a blank carrier (minimum 128×128).
Intent & output
02Verify binding
03After binding, the sealed carrier and binding ID are held client-side. Extract with the same key to recover the intent + output and check the output match. Wrong key fails closed.
Result
04Extracted binding
—
Carrier report
- Binding ID
- —
- Output match
- —
- Bytes embedded
- —
- Capacity remaining
- —
- Encrypted
- —
- Channel
- —
Raw response
—
Operator and workspace consoles are separate applications (pilot).