Claude AI Leads the Way in Antisemitic Content Detection

Artificial intelligence is reshaping how we interact online, but it’s not immune to misuse. From deepfakes to discriminatory content, AI can amplify harmful material if unchecked. A recent study by the Anti-Defamation League (ADL) examined how well leading large language models (LLMs) detect antisemitic content, and the results reveal a clear frontrunner. Claude AI ranked highest, proving the technology can play a key role in combating online hate—while Grok struggled to keep up.

How the ADL Tested LLMs for Antisemitism

The ADL conducted its evaluation between August and October 2025, testing six of the most widely used LLMs. Researchers submitted various types of antisemitic content, including text disguised as survey questions and document summaries, to gauge each model’s detection ability. The study found that LLMs excelled at spotting explicit stereotypes but had a harder time flagging subtler or context-heavy antisemitic material.

Claude and ChatGPT Outperform Competitors

Among all LLMs tested, Claude achieved the highest score, earning an overall performance rating of 80. ChatGPT followed with a score of 57, showing strong but not perfect capabilities. These results suggest that some AI models are better trained or fine-tuned for nuanced content detection, particularly when the material is designed to be misleading or coded.

Where LLMs Still Fall Short

Despite high scores from the top models, the ADL’s report highlighted persistent gaps. Most LLMs struggled with extremist content, particularly anti-Zionist material, which often evades simple keyword detection. Document summaries posed additional challenges, suggesting that AI systems still require contextual understanding to fully identify antisemitic messaging. This underlines the importance of ongoing training and human oversight when using AI for content moderation.

Grok and Llama Lag Behind

On the lower end of the spectrum, Grok scored just 21, placing it behind all competitors, including China’s DeepSeek AI, which scored 50. Llama scored 31, also reflecting significant room for improvement. These findings illustrate that not all AI models are equally capable of handling sensitive or discriminatory material, and users should remain cautious about relying solely on automated moderation.

The Importance of Responsible AI Use

As AI continues to expand into everyday applications, ensuring that systems can detect and prevent harmful content is crucial. The ADL study demonstrates that while some LLMs like Claude show promise, no AI is perfect. Developers, platforms, and users must continue to prioritize ethical AI deployment, combining automated tools with human review to prevent the spread of antisemitic and extremist material.

Improving AI Moderation

The ADL emphasizes that AI models have significant potential but require continuous updates to address blind spots. Enhancing contextual awareness, improving sensitivity to nuanced content, and refining detection of extremist ideologies are key areas for future development. Claude’s performance provides a benchmark, while lower-performing models highlight the urgent need for improvement across the industry.

Claude AI’s top ranking in the ADL study demonstrates the growing potential of AI to combat online hate, but the mixed results of other LLMs, especially Grok, remind us that technology alone cannot eliminate harmful content. Vigilance, ethical development, and ongoing oversight remain essential as AI becomes more embedded in digital communication.