Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
The antisemitism watchdog tested ChatGPT, Claude, Gemini and Grok and found Persian responses hedged or missed antisemitic framing that English responses rejected outright
Getty Images
Illustrative
Major AI chatbots consistently fail to reject antisemitic inquiries in Persian as effectively as they do in English, according to a report released on Wednesday by the Anti-Defamation League. The group argues the safety flaw has broad implications as millions rely on these platforms to understand conflicts such as the U.S.-Israel war against Iran.
The report, “Lost in Translation: How AI Chatbots Fail Persian Speakers on Antisemitism,” looked at responses by the most widely-used chatbots — OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini and xAI’s Grok — over several weeks in March, amid the Iran war.
After assessing eight prompts and 800 total responses, researchers identified a “systemic” bias in how AI models process sensitive topics, heavily dependent on the language used. Across the tested chatbots, English responses proved superior to ones in Persian — a major language with over 110 million speakers primarily in Iran, Afghanistan and Tajikistan — in identifying and rejecting antisemitic conspiracy theories.
Persian-language responses frequently hedged or avoided direct answers where English responses were unequivocal, the ADL found. For instance, Gemini’s English answers consistently acknowledged Iran’s role in spreading antisemitism, while its Persian responses softened or deflected the same claims. Several models also failed to recognize antisemitic framing when questions were posed in Persian. Asked whether U.S. behavior toward Iran had been “Jewlike” — a normalized derogatory term in Persian —- models largely engaged with the question as a neutral political inquiry rather than flagging its antisemitic premise.
The report also identified gaps in depth and sourcing between languages. English responses were generally longer, more thoroughly cited and more up-to-date. Gemini’s English answers referenced the 2026 Iran war and U.S. casualties by name, for example, while its Persian responses only described hypothetical future conflicts. Citations were similarly not consistent. ChatGPT supplied close to 300 links in English responses but none in Persian, and even where other models did cite sources, reliability was inconsistent. Grok cited X accounts including a “Star Trek” fan page and one apparently impersonating Ali Khamenei, the late Iranian supreme leader.
“The gaps we found are not minor inconsistencies, they are systemic failures,” said Daniel
Kelley, senior director of the ADL Center for Technology and Society. “When a platform tells a
Persian speaker that antisemitism is a matter of ‘blurred boundaries’ while telling an English
speaker it is state policy, those are not the same product.”
Kelley called on AI companies “to invest the research and resources necessary to ensure their guardrails work for every person, in every language, and that means starting now.”
While English responses were found to be more conscientious than those in Persian, the ADL raised concerns last year with leading AI developers like Meta and Google regarding antisemitic and anti-Israel biases in English as well. At the time, Meta claimed ADL was using an older model for its study, and not the most current version.
An ADL spokesperson told Jewish Insider that the study on Persian chatbots used the most recent models at the time of the research.
Copyright © 2026 · All Rights Reserved · Jewish Insider
The politics and business news you need to stay up to date, delivered each morning in a must-read newsletter.
Please log in if you already have a subscription, or subscribe to access the latest updates.
Email me a link to sign in
I don’t have an account
Already a subscriber? Sign in
Become a premium subscriber
