Disinformation and synthetic content

Disinformation is false information spread with the intention of misleading. AI makes it cheaper to produce: text, voice and video in quantity, adapted to different audiences.

Signs to look for:

• The picture or video exists in only one place and has no origin.
• No other credible source reports the same thing.
• The content provokes a strong feeling quickly (anger, fear) — it is designed to be shared.
• Details that do not add up: shadows, hands, text in the image, lip sync, background sound.

The best check: a reverse image search, find the original, see what established sources say, wait a few hours.

Watermarking and provenance are technical countermeasures:

• A visible mark («created with AI») — easy to crop away.
• An invisible watermark in the pixels or in the choice of words — it survives some editing, but can often be removed by somebody who wants to.
• Cryptographic provenance (C2PA/Content Credentials) — the camera or the program signs the file with metadata about its origin and edits. The signature cannot be forged, but the absence of a signature proves nothing.

The conclusion is uncomfortable but important: technology can establish authenticity, but not prove falsity. So the weight moves to source criticism: who says it, why, and what do independent sources say?

The EU AI Act requires synthetic content to be labelled — that raises the bar but does not solve the problem.