ADL Finds AI Chatbots Miss Antisemitism in Persian

ADL's Center for Technology and Society said its new study found major AI chatbots were weaker at recognizing and rejecting antisemitic prompts in Persian than in English, a gap exposed during the 2026 Iran war. Researchers tested ChatGPT, Gemini, Claude and Grok across eight prompts and 800 responses from March 9 to 30, including questions about Zionist conspiracy tropes, Iran and the United States. ADL said Persian answers often softened or missed antisemitic framing, while English answers were more complete and grounded. The finding matters for Israel and Jewish communities because adversaries weaponize Persian-language disinformation, and AI guardrails must protect users in every language.