Deepfake scams: authenticate instructions beyond the voice or video

Synthetic media changes the trust equation

Voice imitation can make an urgent instruction sound like it came from a manager or relative. Synthetic video may also create a convincing but entirely fabricated encounter.

Such material does not mean all video calls are unsafe. It means that recognisable appearance alone should not authorise sensitive financial or account actions.

Second-channel checks remain effective

Confirm unexpected requests using a separate channel or previously agreed procedure. Avoid relying on a callback number supplied during the suspicious contact.

Automated deepfake detection may miss manipulated content or falsely flag authentic material. Results should be assessed alongside context and provenance.

Practical checks

• Require independent approval for unusual transfers
• Maintain verified contact details
• Review detector outputs critically
• Avoid unnecessary sharing of sensitive recordings

Technical detectors should not be the final authority

Organisations can practise scenarios involving impersonated executives and ensure payment approval rules apply equally to audio, video and written instructions.

Alain Stevens’s work with online information and fraud-related risks since 1999 supports an approach based on verifiable sources, especially as generated media becomes more accessible.

Related resources

Build a suitable control

The strongest defence is a process that can decline an apparently familiar instruction when it fails verification.

For a technical question, software project or training request, describe the need without sending credentials or unnecessary personal data.