For most of history, a photograph or a recording was strong evidence that something happened. That assumption is ending. Generative AI can produce realistic images of events that never occurred, clone a voice from a short sample and synthesise video of people saying things they never said.

Two dangers, not one

The first danger is obvious: fabricated media presented as real. A fake audio clip of an official, a generated photo of a disaster, a doctored video of a public figure.

The second danger is quieter and just as serious. Once people know that fakes are possible, real evidence can be dismissed as fake. Researchers call this the "liar's dividend": anyone caught on a genuine recording can claim it was generated. Synthetic media erodes trust in authentic media too.

Why detection alone will not save us

Tools that detect generated content are useful, but they are in a constant race with the tools that generate it. Every improvement in detection is followed by improvements in generation. A detector that is right most of the time will still be wrong often at internet scale.

Provenance: proving where something came from

A more durable approach is provenance: recording, at the moment of capture, who created a piece of media and how it was edited. The Coalition for Content Provenance and Authenticity (C2PA) publishes an open technical standard for attaching this kind of tamper-evident history to images and video. Instead of asking "does this look fake?", provenance asks "can this prove where it came from?"

Habits that still work

  • Check the source, not the pixels. Who first published the file, and do they have a track record?
  • Look for independent confirmation. A real event is usually recorded by several people from different angles.
  • Be wary of perfect timing. Media that appears exactly when it is most politically or emotionally useful deserves extra scrutiny.
  • Slow down on audio. Voice cloning is cheap and convincing; confirm unusual requests through another channel.

Where Honest Lens stands

Honest Lens analyses text claims and news articles; it does not claim to detect synthetic images or audio, and it will not pretend to. What it can do is check the claims that travel with such media. A fabricated image usually arrives with a caption, and the caption makes checkable claims: who, where, when, how many. The Claim to Evidence engine can test those.