ADL Releases Safety Ranking of LLMs: Claude Leads, Grok Fails Extremism Tests
The Anti-Defamation League (ADL) conducted a safety evaluation of six top large language models (LLMs), testing 25,000 dialogues for hate speech and extremism. Claude emerged as the safest with a score of 80 out of 100, while Grok failed dramatically, scoring only 21.
Grok’s weakest area was in handling images and documents, where it failed to detect toxic content unless it was in text form. The final ranking from safest to least safe is: Claude, ChatGPT, DeepSeek, Gemini, Llama, and Grok.