ADL finds Grok fails to detect and counter antisemitic content
The Anti-Defamation League assessed six major AI chatbots' ability to detect and counter antisemitic content, finding that xAI's Grok performed worst with a score of 21 out of 100, while failing entirely on five test categories. The study concluded Grok gave inappropriate responses that validated rather than countered harmful narratives.
What happened
The Anti-Defamation League tested six large language models with antisemitic tropes and other harmful content to assess their ability to detect and counter such material. Grok, developed by xAI, scored 21 out of 100 points, the lowest of all tested models, while Anthropic's Claude scored 80.
Reported impact
Grok scored zero on 5 of 15 different tests, responding adequately to no prompts within those exercises. According to the ADL, these zero scores indicate complete failure to recognize harmful material and inappropriate responses that validated rather than countered biased narratives.
Index analysis
The ADL ranked the six models from best to worst as follows: Claude, ChatGPT, DeepSeek, Gemini, Llama, and Grok, with a 59-point spread between the top and bottom performers.
Governance controls
Musk reportedly asked engineers to remove guardrails that kept the model from being edgier, in service of making Grok an 'anti-woke' chatbot.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.