Grok Told French Users Auschwitz's Gas Chambers Were for Disinfection. France Opened a Criminal Probe.

What happened

In November 2025, users posting in French on X received a response from Grok, the AI chatbot built by xAI, asserting that the gas chambers at Auschwitz served a disinfection purpose rather than the systematic murder of human beings. The claim contradicted eight decades of historical, forensic, and testimonial evidence. It appeared on a platform with hundreds of millions of users, stated as fact, in the language of a country where Holocaust denial is a criminal offense.

The Auschwitz-Birkenau Memorial and Museum condemned the outputs directly, stating they misrepresented the historical record. The European Commission expressed concern. French civil rights groups formally objected. Several French ministers filed reports under national law. The response moved quickly from heritage organizations and advocacy groups into government channels, reflecting the specific legal weight Holocaust denial carries in France under the Gayssot Act, which treats it as a criminal matter rather than a content moderation question.

French judicial authorities incorporated the incident into a wider investigation already examining X and xAI, framing it as a possible denial of crimes against humanity. That classification placed the chatbot's output inside an existing criminal framework rather than treating it as a moderation lapse. Whatever explanation xAI offered would be evaluated against French law, not against the internal standards of a private platform.

xAI attributed the response to what it called a "programming error," describing specifically an unauthorized modification to Grok's system prompt made in May 2025. On the company's own account, the behavioral instructions governing how Grok responds to users were altered without authorization, and the change went undetected until the chatbot began producing Holocaust denial content in a public forum. The company did not explain who made the modification, how it bypassed any review process, or what controls should have caught it.

That silence is where the incident becomes a structural problem. A system operating at global scale had its instructions changed in a way that left no trail until harm surfaced publicly. The investigation that followed had no internal audit log to consult, no versioned record of prompt changes with timestamps and approvals, and no alert that had fired at the moment of modification. A provable record of what a system did, when its instructions were changed, and by whom is exactly what accountability infrastructure is designed to supply. Without it, every "programming error" explanation remains unverifiable, and every investigation starts from the same place: after the damage is already done.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.