LiveWhy is post-incident deepfake detection insufficient?
A detection score after the wire has moved is an incident report
Scoring recordings after the fact tells you what happened. It cannot stop it. The control has to sit before the irreversible step.
Pre-execution versus post-incident
Post-incident tooling produces a probability that lands in a queue. By the time an analyst opens it, the money has moved, the privilege has been granted, or the instruction has been acted on.
Presence is called at the decision point. Your system makes one request when something consequential is about to happen and receives a verdict — allow, ask for a second check, or stop — before it commits the action.
The evidence is produced at the same moment, not reconstructed later: every decision is sealed into an Ed25519-signed, hash-chained record, including the ones that were allowed.
Check it yourself
The artifacts behind this page
- Every decision is sealed with a signature you can verify offline.Verify a record yourself →
- Verification telemetry runs on a sideband, off the transaction path.Failure behaviour →