What happened

Directly following the 10th anniversary commemorations of the terror attacks on Paris' Bataclan theatre, in which Islamic State militants killed 90 people during a concert on November 13, 2015, the AI chatbot surfaced conspiracy theories claiming there were "crisis actors" and that the attacks had been staged. 

Grok also repeated fabricated claims about torture and mutilation that have been repeatedly disproven by French investigators. 

The false claims echoed narratives spread at the time by far-right groups to falsely undermine media and official reports of the tragedy. 

The claims surfaced when users posted screenshots showing Grok supplying the disinformation in response to queries about the attacks. 

Why it happened

The failure stems from well-known issues with large language models trained on unvetted internet content, particularly when the parent platform (X) hosts large volumes of extremist misinformation.

Grok’s design emphasises a “rebellious” or “edgy” personality with fewer restrictions on what can be published, which further reduces its safeguards. 

Limited transparency about its training data, safety filters, and reinforcement processes, combined with an apparent lack of fact-checking or domain-expert review, likely allowed debunked conspiracy theories about the Bataclan attacks to propagate through the model.