The problem
Early in 2024, a finance employee at the engineering firm Arup wired $25.6 million during a video call in which every colleague on screen was synthetic. DISCERN-AI was built to stop that call.
Most detectors chase the fingerprints a particular generator leaves behind, so every new release can break them.
The approach
DISCERN-AI measures something else: the closed-loop signature of a living body, where breath, voice and motion keep adjusting to one another. A generator can copy how a person looks and sounds. The question DISCERN-AI asks is whether that loop is there.
The verified numbers
Benchmark edition of 9 November 2026. Every number is from a complete run.
Voice. 13 AI voice engines, none of them seen in training. Eleven are caught at 95.9% or better.
Video. Each generator is held out of training in turn and then tested, so every score is on a generator the model has never seen.
Full results document (PDF) → Charts for every engine and generator →
Where it applies
- Financial fraud
- Call-centre authentication
- Insurance investigation
- Legal and forensic use
- Media verification
- Identity platforms
Intellectual property and licensing
Protected by U.S. Non-Provisional Patent Application 19/700,318, pending (priority 13 June 2025).
One exclusive partner per vertical. All rights outside that vertical are retained.
Read more
Licensing & collaboration
Offered for licensing, not for sale. Independent validation, engineering collaboration and research partnerships are welcome.
[email protected]Inventor: Paul Samuel Guarino, Independent Researcher & Inventor