Skip to main content
Every test call is also a test of DeepBlocker’s deepfake detector. After the call, the detector scores the attacker’s voice on the recording and says whether it would have flagged it. The results page puts that verdict next to the person’s verdict, on the same call. The console explains the column like this: “Detector caught: your blue-team deepfake detector flagged the synthetic voice, out of the calls it scored. Higher is better. ‘Not scored’ means the detector has not finished analysing yet (it scores shortly after the call).”

​
The three states

Hover a badge for the detail: Scored X% fake (threshold Y%), and the model name when known. A failed scoring shows Scoring failed with the reason.

​
The threshold

The threshold is 80% unless the detector reports another for that call. A call scoring 80% fake or more is Caught. The hover text always shows the threshold that was applied, so a verdict can be checked.

​
Where it appears

​
Why it matters

The worst case is a person who failed on a call the detector also missed. The results page calls this Got through, or Residual. You want it at 0. A high catch rate on the calls your people fell for means a real attack with that voice would have been stopped before anyone picked up.
On a call-back trap the caller is a real member of staff, not the AI. There the detector scores a real voice, and a real employee should not be flagged. A Caught on that view is a false alarm worth telling us about.

​
Problems

The detector scores a call shortly after it ends. Wait a few minutes, then click Sync & analyze on the results card. A call with no usable recording stays Not scored.
Verdicts land as recordings are scored. The percentages are worked out over the calls scored so far, so they settle once every call has a verdict.
The detector reported its own threshold for that call. The verdict uses the threshold shown.