Grok: the most "politically incorrect" AI, according to the ADL (and why that's no surprise)

Elon Musk has always promised us an AI that tells the "truth," however uncomfortable it may be. The Anti-Defamation League (ADL) seems to have taken that promise literally and set out to test where the line between honesty and outright hatred lies. The results of the new study are exactly what skeptics expected: xAI's Grok has been officially named the most problematic chatbot on antisemitism. While its competitors build digital fences, Grok seems to invite anyone willing to come in and debate the slipperiest topics.

The ADL researchers did not simply ask blunt questions. They used a sophisticated system of prompts split into three categories: anti-Jewish statements, anti-Zionism and extremism. All the industry heavyweights took part, from ChatGPT and Gemini to Llama and China's DeepSeek. Anthropic's Claude proved a model student, filtering out almost every hint of toxicity with near-sterile thoroughness, while Musk's creation showed a remarkable flexibility, in the worst sense of the word. Grok did not merely let dubious claims through; it often picked them up, becoming a digital mouthpiece for ideas that polite society prefers to avoid.

It is important to understand that this failure is no accident. It is not a bug in the system but rather its key feature. When Musk launched xAI, he positioned Grok as the antithesis of the "sterile" and "woke" models from OpenAI and Google. The irony is that, in trying to avoid censorship, Grok's developers built a product that ignores basic safeguards. In a world where AI is becoming the main source of information for millions, that stance looks risky at best. While Google Gemini apologizes for every trifle, Grok charges ahead, not much caring whose feelings it hurts along the way.

The other players are not perfect either. Even Claude, which took first place, has gaps that need work, according to the ADL. This highlights a fundamental problem for the industry: we still haven't learned to teach machines ethics in a way that doesn't turn into censorship or, conversely, into indulging hatred. OpenAI and Meta* try to strike a balance, but their models often get confused by context, blocking harmless historical references while letting veiled insults through. Against this backdrop, Grok looks like a rebellious teenager who deliberately does the opposite just to annoy the teachers.

The consequences of this report for xAI may be far more serious than bad press. Advertisers and corporate clients are the people who keep tech giants alive. A major brand is unlikely to want to build into its workflows a tool that spouts extremist slogans every other time. Musk can of course keep pushing his line of absolute free speech, but the market dictates its own rules. If Grok doesn't learn to watch its language, it risks remaining a toy for a narrow circle of ideological supporters, losing its chance at mass adoption in the business sector.

The Grok situation raises a global question: who should set the limits of what neural networks may do? If we leave it to corporations, we get a sterile, useless AI. If we leave things as they are at xAI, we get a toxicity generator. The truth, as always, lies somewhere in between, but for now Grok sits at the extreme end of that scale. The ADL study was a cold shower for those who believed AI could exist outside politics and social norms.

The bottom line: Grok deliberately ignores ethical filters to preserve its image as an honest AI, but in the eyes of large organizations that turns it from an innovation into a reputational landmine. Can Musk balance free speech with safety requirements, or will Grok remain the most toxic chatbot of our time?

*Meta is designated an extremist organization and banned in Russia.