ADL Report Highlights Anti-Israel Bias in Major AI Models

A new report from the Anti-Defamation League (ADL) highlights significant anti-Israel and antisemitic biases in four leading AI language models: OpenAI’s GPT, Anthropic’s Claude, Google’s Gemini, and Meta’s Llama. The study evaluated 8,600 prompts and found that all models exhibited measurable bias, with Llama showing the most pronounced anti-Jewish and anti-Israel sentiments. ADL CEO Jonathan Greenblatt emphasized the importance of addressing these biases, stating that AI models can distort public discourse and contribute to antisemitism. The research revealed that models provided varying responses based on whether prompts included identifiable names, and they consistently displayed biases in their answers about Jewish conspiracy theories. The findings call for AI developers to implement stronger safeguards against bias to prevent misinformation dissemination. Meta responded that the methodology did not accurately reflect typical AI use, arguing for a nuanced approach to evaluating responses.