Grok posts antisemitic content for 16 hours after a code update
After an update on July 7, Grok posted antisemitic content on X. xAI apologised and said deprecated instructions made the bot echo extremist posts.
- Occurred
- Disclosed
- Published
- Organization
- xAI
- System
- Grok response bot on X
- Impact
- Public chatbot posted antisemitic and extremist content for about 16 hours
- Root cause
- Update to an upstream code path that reintroduced deprecated instructions, according to xAI
Timeline
- 2025-07-07, about 11 PM PT. The update is deployed, according to xAI.
- Following 16 hours. The changed code is live.
- 2025-07-12. Both reports cover xAI’s apology.
Trigger or entry point
An update to an upstream code path for the @grok account.
Failure mechanism
According to xAI’s statement, the update added deprecated instructions that made Grok susceptible to existing posts on X, including posts with extremist views. xAI stated that the root cause was independent of the underlying language model. Engadget places the incident on July 8, the day after the update.
Impact
Grok posted antisemitic content publicly on X. Engadget reports that the chatbot was disabled temporarily. The posts were public replies on X, visible to anyone, during the 16 hours the code was live.
Response
xAI apologised, removed the deprecated code and said it had refactored the system. The two reports differ on when the apology was posted: NBC News says Saturday, Engadget says late Friday night. Both articles are dated July 12. The statement was posted from the Grok account on X, and both reports link to it.
Mitigations
xAI said it would publish the new system prompt in its public GitHub repository. Neither report describes a change to how updates to the bot are reviewed before release.
Sources
- Grok statement on Xprimary · 2025-07-12
- AI chatbot Grok issues apology for antisemitic postssecondary · 2025-07-12
- Grok team apologizes for the chatbot’s ‘horrific behavior’ and blames ‘MechaHitler’ on a bad updatesecondary · 2025-07-12
Corrections
No corrections have been made to this entry.
Report a correction