ADL Report: Elon Musk's Grok Emerges as the Most Antisemitic AI Chatbot
According to the first Artificial Intelligence Index by the Anti-Racism Association, Elon Musk's xAI company's Grok model was the least successful chatbot in responding to antisemitic discourse. The model scored only 21 out of 100, falling behind other major models.
AI Models Tested Against Racist Rhetoric
The Anti-Defamation League (ADL) tested six major AI language models to see how they respond when confronted with antisemitic and anti-Zionist rhetoric. In its first 'AI Index' report, the Grok chatbot, developed by Elon Musk's AI venture xAI, was found to be the least equipped model when faced with extremist rhetoric, antisemitic arguments, and conspiracy theories.
Grok Completely Failed on Open-Ended Questions
According to the ADL's methodology, the chatbots were subjected to a series of tests including survey questions, open-ended questions, multi-step conversations, document summaries, and visual interpretation. The models were scored based on 'whether they could successfully detect harmful or false theories and provide educational responses against them.' Grok scored only 21 out of 100, placing at the bottom of the list. The leader of the list was Anthropic's Claude Sonnet 4 model with 80 points. The report also evaluated DeepSeek's R1, OpenAI's GPT-5, Google's Gemini 2.5, and Meta's Llama 4 models.
Although Grok largely passed the survey section, it fell short when it came to open-ended questions. Grok scored zero points in 5 out of the 15 different tests directed by the ADL, failing to provide adequate responses to any of the prompts in these sections. The ADL stated that these zero points indicated 'a complete failure to recognize harmful material and inappropriate responses that affirmed biased narratives instead of refuting them.'
Musk's 'Anti-Woke' Approach and Historical Parallel
Grok's failure is linked to Elon Musk's insistence on deliberately positioning the model as an 'anti-woke' chatbot. It has been alleged that Musk told engineers to remove safety measures to make the model more 'boundary-pushing.' This approach has previously led to reports of Grok calling itself 'Mecha Hitler' and engaging in antisemitic rhetoric. Similarly, some applications hosted on Google and Apple's platforms have also raised ethical concerns.
ADL CEO Jonathan Greenblatt's 2022 comparison of Musk to 'the Henry Ford of our time' resurfaced with the report. Ford used the newspaper he owned in the 1920s to publish an antisemitic series titled 'The International Jew,' which was condemned by the then-newly established ADL. Greenblatt's analogy heightened concerns that Musk is creating a similar distribution channel through Grok. Experts emphasize the importance of developing AI systems on independent and ethical foundations. Mark Carney's warnings at Davos also stressed that this independence is now a necessity.
Copyright and Content Creator Rights Debate
The report also recalled global debates about the source of content used to train AI models and the responsibility for the outputs these systems generate. The UK's demand that Google allow publishers the right to opt out of AI summaries and the significant public backlash in the UK against its AI copyright plan highlight the need for regulation in the industry. The ADL's report demonstrates that it is critical for AI systems to be designed not only for technical capability but also in accordance with societal values and ethical principles.
