What happened

Following the stabbing incident on a train from Doncaster to London (stopped in Huntingdon, Cambridgeshire) on 1 November, Grok told users that a widely shared image showing a wounded man lying in a train carriage, with paramedics and police around “appears to be a genuine photo.”

However, multiple red flags suggest the image is AI-generated: the text on officers’ clothing is garbled; there’s a stylised filter; and the seating in the train doesn’t match the real train involved.

An X (formerly Twitter) account that circulated the photo even appeared to confirm it was AI-generated.

Google Lens’s AI overview likewise mischaracterised the image. In one search, it claimed it was “a still from a BBC News report” and linked to a BBC article — but that story doesn’t actually include the image.

On separate occasions, Google Lens’s AI linked a second piece of video (showing a train confrontation) to the Huntingdon attack — but Full Fact argues this video almost certainly isn’t real footage of the incident.

Why it happened

The issue stems from limitations in how these AI systems detect and classify generated content. Grok, for example, has a documented history of misidentifying AI-generated images or videos as real.

There may also be systemic vulnerability: if an AI is trained on broad internet data where disinformation or manipulated media proliferates, it can reinforce and repeat false narratives. Full Fact notes such errors are not new. 

On the corporate side, accountability is limited: while users can flag Google Lens AI errors (e.g., via “thumbs down”), more robust validation or oversight mechanisms don’t appear to be working well enough.