Deepfakes and the collapsing presumption that seeing is believing

The liar’s dividend

Deepfakes create obvious risks of impersonation, fraud, reputational harm and fabricated admissions. A second risk is less visible: a party may deny genuine audio or video simply because fabrication is technologically possible. This ‘liar’s dividend’ makes provenance and collection discipline more important for all media, not only content suspected of being synthetic.

Detection tools provide indicators, not infallible verdicts. Compression, editing, screen recording and platform processing can degrade the traces on which those tools rely. Legal conclusions should not be based on a single automated score.

Provenance, context and corroboration

A robust evidence file preserves the original file, metadata, device and account records, transmission history and witness evidence explaining how the content was created and obtained. Experts may examine visual artefacts, audio characteristics, encoding history and consistency with known source material, but their limitations should be stated.

Context may be as important as technical analysis. Contemporaneous messages, location data, transaction records and witness conduct can corroborate or contradict the media. Courts ultimately evaluate the whole evidential picture under the applicable rules of proof.

Governance and disclosure

Organisations producing synthetic content should maintain provenance records and clear approvals. Current regulatory approaches increasingly require machine-readable marking or disclosure for certain AI-generated or manipulated content. Even where a specific rule does not apply, transparent labelling can reduce deception risk and preserve trust.