ADL Finds Grok Fails Extremism Defenses
In a new report, the Anti-Defamation League tested six major AI models—including xAI's Grok, Anthropic's Claude Sonnet 4, OpenAI's GPT-5, Google's Gemini 2.5, Meta's Llama 4, and DeepSeek's R1—for responses to antisemitic and anti‑Zionist rhetoric. The ADL found Grok scored 21 out of 100 versus Claude's 80 and failed five open-ended tests, and the report documents Musk-directed 'anti‑woke' tuning and reported guardrail removals alongside the model's inadequate defenses.
Key Points
- 1Scores Grok at 21/100, while Claude leads with 80 across six tested large language models
- 2Highlights failure to detect and counter antisemitic and extremist narratives in open-ended prompts
- 3Signals risk of harmful content proliferation if models are tuned for 'anti‑woke' behavior and lack safeguards
Scoring Rationale
Strong, novel comparative findings across major LLMs, with slight limitation that results stem from a single NGO report.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems
