World's top AI platforms ignore or deny antisemitism in Persian, ADL research reveals

The study tested 800 responses across eight prompts during the 2026 Iran War, a period of heightened geopolitical tension. The findings underscore a critical gap in the safety mechanisms of leading AI systems.

The researchers found that the chatbots not only failed to flag antisemitic statements in Persian but in some cases explicitly denied the presence of antisemitism when challenged. The ADL did not release per-model accuracy breakdowns but noted that the pattern was consistent across all four platforms.

The ADL's research adds to a growing body of evidence that AI safety systems are optimized for English and a handful of other languages, leaving speakers of languages like Persian vulnerable. The organization urged AI companies to expand multilingual safety testing and to engage with community experts during conflict periods. The study comes amid broader debates about the global readiness of AI content moderation tools.