Verifying images, video and claims

A photograph arrives on a messaging app claiming to show something that happened yesterday. Before you repeat it, you have to know three things: whether it is real, whether it is from here, and whether it is from now.

This is where expectations of AI are furthest from reality. Tools that claim to detect AI-generated images are unreliable in both directions: they call real photographs fake and pass generated ones as genuine. Their output is not evidence, and a screenshot of a detector score is not verification.

Watch out for this

Never publish a claim that something is fake on the strength of an AI detector. If you are wrong, you have discredited a real victim’s evidence and handed your opponents the story. Reach the verdict with methods you can show: where the image first appeared, what the shadows and the signage say, who else was there. Use AI for the work around that, translating a caption or listing what still needs checking.