Harmful outputPI-0015
xAI's Grok posted antisemitic content and praise of Hitler on X after a system-prompt change
For about 16 hours, Grok's X replies repeated antisemitic tropes and praised Adolf Hitler; one attached a Jewish-sounding surname to a photo of an unrelated woman. The company deleted the posts, froze the account's replies, apologized and said a code update had made Grok draw on extremist user posts and follow instructions to be provocative.
Harm: Impact unknownControl level 1, No rule broken
Not countedEffect on the index
Sources
How we know
6 sources · developer confirmed. Links go to the original publishers; the summary above is in our own words.
- primaryGrok account statement: xAI is removing the inappropriate posts and banning hate speech before Grok postsxAI (@grok on X) · July 8, 2025x.com/grok/status/1942720721026699451
- primaryGrok account apology for the horrific behavior, blaming a code-path updatexAI (@grok on X) · July 12, 2025x.com/grok/status/1943916977481036128
- newsMusk pulls hateful X posts after AI chatbot lauds Hitler, curses Polish PMPolskie Radio · July 9, 2025polskieradio.pl/395/7789/Artykul/3548854,musk-pulls-hateful-x-posts-after-ai-ch…
- newsCalling itself 'MechaHitler,' Elon Musk's AI tool spreads antisemitic conspiraciesThe Forward · July 8, 2025forward.com/fast-forward/753703/grok-cindy-steinberg-elon-musk-antisemitism-ai…
- newsElon Musk's AI chatbot Grok gets an update and starts sharing antisemitic postsAP (via Spectrum News 13) · July 10, 2025mynews13.com/fl/orlando/ap-top-news/2025/07/09/elon-musks-ai-chatbot-grok-gets…
- newsAI chatbot Grok issues apology for antisemitic postsNBC News (via NBC Bay Area) · July 12, 2025nbcbayarea.com/news/national-international/ai-chatbot-grok-issues-apology-for-a…
Why this rating
Impact unknown; control failure level 1
Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.
Observed harm
Impact unknownHateful posts were public for about 16 hours, but no harm to an identified person is documented: the post naming a surname attached it to a photo of an unrelated woman. Harassment of a named target (Will Stancil) was requested by a user, which is deliberate misuse under section 0 and not rated here. xAI's apology refers to distress that many people experienced, not to a bounded harm.
Unknown is not the same as none.
The harm scale
- 1 Negligible Inconvenience, easily remedied.
- 2 Minor Limited, recoverable harm.
- 3 Moderate Material harm needing significant effort to remedy.
- 4 Severe Severe harm to health, rights, property or essential services.
- 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.
Control assessment
No rule brokenThe model behaved as its developer's changed instructions steered it; xAI attributed it to its own code update.
No rule was broken, so it cannot set the control floor.
The control scale
- 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
- 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
- 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
- 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
- 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.
Rating rationale
Impact unknown: the hateful outputs are confirmed by xAI, but no source documents a bounded resulting harm to identified people; reach and views are not counts of people harmed. User-requested harassment (for example of Will Stancil) is deliberate misuse and out of scope. Control level 1: the model followed its developer's changed instructions. The behavior label is a stretch (harmful output rather than guidance).
Effect on the index
It does not count toward the July 14 reading
The reading for the week to July 14, 2025, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.
Not counted: impact unknown. 3 other records behind the reading for that week.
The arithmetic
| Step | With it | Without |
|---|---|---|
| Counts toward the index?documented, external, eligible evidence | No | — |
| Worst documented harm, ksets the band | — (none counting) | — (none counting) |
| Harms at that level, nposition in the band | 0 | 0 |
| Highest control level breachedsets the reading only when no harm counts | 2 | 2 |
| Readingrounded down | 2 Control failures only | 2 Control failures only |
Inside the window of 4 weekly readings
| Week to | Reading | Band |
|---|---|---|
| July 14, 2025 | 2 | Control failures only |
| July 21, 2025 | 9 | Negligible harm |
| July 28, 2025 | 9 | Negligible harm |
| Aug. 4, 2025 | 3 | Control failures only |
Revisions
What we changed
2 logged. Every change to a rating is logged here, with the reason.
- v2Oct. 13, 2026
Wording only: rubric shorthand in the rating rationale replaced with plain words; ratings unchanged.
- v1Oct. 7, 2026
First published.
Cite and share
Use this record
Citation
Paperclip Index. “xAI's Grok posted antisemitic content and praise of Hitler on X after a system-prompt change.” Record PI-0015. Reported July 8, 2025; updated Oct. 13, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0015