A new report from the Anti-Defamation League (ADL) reveals alarming trends in the misuse of large language models (LLMs) like Google’s Gemma-3 and Microsoft’s Phi-4 by online extremists promoting antisemitism and genocide. The ADL’s study, “The Safety Divide: Open-Source AI Models Fall Short on Guardrails for Antisemitic, Dangerous Content,” examined 17 LLMs, finding that open-source models pose a critical vulnerability due to their lack of safety measures. Notably, OpenAI’s GPT-4o outperformed all open-source models, but GPT-5 showed higher risks with more harmful content and lower refusal rates. The findings emphasize the need for stringent controls, safety audits, and collaboration with civil society to mitigate the risks of AI being weaponized for hate speech. Additionally, researchers highlight that extremist groups are increasingly utilizing AI for propaganda, suggesting a shift in digital tactics. The ADL and MEMRI’s reports underline a pressing need for regulatory action to protect vulnerable communities.
Source link
