Chapter 5. Deepfakes, Disinformation, and the Crisis of Trust

5.1 Synthetic media changes the evidence environment

Generative systems can create or alter text, images, audio, and video. This can support legitimate creativity, accessibility, translation, training, and entertainment. It can also produce impersonation, fabricated endorsements, false public statements, manipulated evidence, and misleading news-like material. A deepfake is synthetic or manipulated media that makes a person appear to say or do something that did not happen.

The risk is not limited to convincing fakes. The existence of synthetic media can create a “liar’s dividend”: a person may dismiss genuine evidence by claiming it is fake. Trust is harmed both when false material is believed and when authentic material becomes easier to deny.

5.2 How to evaluate suspicious content

Pause before sharing emotionally intense, urgent, or shocking content.
Check whether the original source is identifiable and whether reputable outlets or official channels independently confirm it.
Look for context: date, location, full recording, preceding and following statements, and whether the clip has been edited.
Contact the person or organisation through a known, independently obtained channel when impersonation is suspected.
Treat visual artefacts and AI detectors as clues, not proof. Detection tools can produce false positives and false negatives.
Preserve relevant evidence safely if reporting fraud or abuse, and avoid reposting harmful material unnecessarily.

5.3 Watermarks, provenance, and disclosure

Technical provenance systems and machine-readable labels can help establish how a file was created or edited, but they are not complete solutions. Metadata can be removed, content can be re-recorded, and a label does not establish that a claim is truthful. Human-readable disclosure, trustworthy distribution channels, verification practices, and accountability must work together.

As of this 2026 edition, the European Union’s AI Act transparency provisions under Article 50 include obligations concerning certain AI interactions and AI-generated or manipulated content, with the European Commission stating that these transparency obligations apply from 2 August 2026. Legal scope and exceptions are specific, so organisations must consult the current official text and qualified counsel rather than generalising EU rules to every country (European Commission, 2026).

5.4 Protecting reputation and community trust

Before publishing synthetic media, ask whether a reasonable viewer could mistake it for authentic evidence, whether a real person’s likeness or voice is used with permission, and whether the content could mislead, humiliate, or endanger someone. Label realistic synthetic content clearly. Do not fabricate a real person’s endorsement or make a person appear to have made a statement without consent.

Institutions should establish a fast, fair response process for impersonation incidents. Correct false claims visibly, avoid amplifying them in the correction, and communicate what has been verified rather than overstating certainty.