isGenAI.com

Glossary

Deepfake: the legal definition and the file evidence

A deepfake, as defined in the EU AI Act, is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful.

Definitions and isGenAI behaviour as of October 2026. Updated .

See the term in a real reportFrequently asked questions
Code-drawn illustration for Deepfake

Illustration · PNG file

  1. Signature + IHDR
  2. iTXt XML:com.adobe.xmprecord
  3. tEXt parameters
  4. tEXt prompt / workflow
  5. tEXt invokeai_metadata
  6. eXIf
  7. caBX C2PA manifest
  8. IDAT pixels
  9. IEND

AI source type declared

The file labels itself as AI-made.

IPTC DigitalSourceType such as trainedAlgorithmicMedia is read from XMP and shown with its exact value.

Check an image →
AI source type declared: where this record sits inside a PNG file and what the isGenAI report says about it. The picture is a code-drawn illustration.

The legal definition and duty

Article 3 of the EU AI Act defines the term, and Article 50(4) requires deployers who generate or manipulate a deepfake to disclose that the content has been artificially generated or manipulated, with narrow exceptions for evidently artistic or satirical works, which need only a disclosure that does not hamper the work. The disclosure duty applies from 2 August 2026. Article 50(2) separately requires providers to mark outputs in a machine-readable way.

The law regulates disclosure. It does not define a detection method.

How isGenAI helps with a suspected deepfake

isGenAI reads what the file records: a C2PA manifest naming an AI tool, an IPTC source type, an AI statement, a saved prompt or a matching copy of a known original. It reports each with its source and does not produce a deepfake probability from the pixels. For video, the C2PA checker validates MP4 and MOV manifests and the frame extractor produces stills for matching. The word deepfake found in metadata is quoted as an AI statement.

What to do with a stripped copy

Most deepfakes circulate as re-encoded copies with no records. Trace the first publication, obtain the original where possible, and compare copies by fingerprint. Platform labels on TikTok, YouTube or LinkedIn are observations worth recording beside the file report.

How to use this term when checking a file

  1. Upload the original file to the AI image detector or the tool named on the page.
  2. Find the finding that uses this term under “How we checked” and read the source field it names.
  3. Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
  4. Remember that absence of a record is not evidence of a human author.

See the term in a real report

Frequently asked questions

What is the EU AI Act definition of a deepfake?

AI-generated or manipulated image, audio or video that resembles real persons, objects, places, entities or events and would falsely appear authentic or truthful.

Who must label a deepfake?

Under Article 50(4), the deployer who generates or manipulates it, from 2 August 2026, with limited artistic and satirical exceptions.

Can isGenAI tell me if a video is a deepfake?

It reports the file’s provenance records and matching copies. It does not score pixels, so a stripped copy returns no evidence found.

What is the best evidence against a deepfake claim?

A valid, trusted camera credential in the original file, or a matching original with such a record.

Sources

Definitions follow the published standards linked above; the isGenAI behaviour described is the deployed checker as of October 2026.