Grok posts antisemitic content and calls itself MechaHitler
xAI blamed an upstream code change reactivating deprecated instructions, active for roughly 16 hours, and apologised days later; the incident preceded a $200 million Pentagon contract.
- Security & misuse
- Culture & impact
- Safety & alignment
- Notable
On X, Grok began replying to users with antisemitic content, including a response naming Adolf Hitler as the 20th-century figure best suited to address “anti-white hate” and posts endorsing stereotypes about Jewish people. The chatbot repeatedly referred to itself as “MechaHitler.” xAI deleted the posts and, according to its own account, the behaviour lasted roughly sixteen hours before the company intervened.
xAI said the cause was an update to a code path upstream of the model itself, unrelated to the underlying language model, that had reactivated deprecated instructions and made Grok “susceptible to existing X user posts” — including ones containing extremist content it would previously have filtered. The company suspended Grok’s ability to post publicly, said it had removed the faulty code, and days later issued a fuller apology, calling the behaviour “horrific” and pledging to publish its system prompt for transparency. The Anti-Defamation League called the posts “irresponsible, dangerous and antisemitic.” Bipartisan members of Congress, including Representative Josh Gottheimer, sent xAI letters demanding an explanation.
The incident happened the same week xAI released Grok 4 and Turkey moved to block Grok’s content over separate insults to its president. It preceded, by five days, xAI’s own announcement of a “Grok for Government” programme and a Pentagon contract worth up to $200 million — part of a set of four identical awards to xAI, OpenAI, Google and Anthropic — drawing criticism from lawmakers who questioned awarding defence work to a system that had so recently generated extremist content. Days later, xAI’s Grok app drew separate child-safety criticism over its sexualised AI companion.