Grok Scores Lowest in ADL Safety Audit

xAI's Grok scored 21 out of 100 in the Anti-Defamation League's comprehensive AI antisemitism detection study, ranking last among six major large language models tested. This represents a devastating 59-point gap behind the top performer, indicating systemic failures in Grok's content moderation and hate speech detection capabilities.

Claude Leads with Highest Safety Score

Anthropic's Claude topped the rankings with 80 out of 100, demonstrating significantly better performance at identifying and countering antisemitic content compared to competitors. However, the ADL cautioned that even Claude and other high-performing models showed dangerous gaps that require immediate attention and remediation.

Comprehensive Testing Methodology Across 25,000+ Prompts

ADL conducted rigorous testing with over 25,000 prompts across three critical categories "anti-Jewish," "anti-Zionist," and "extremist" between August and October 2025. The study evaluated six major models: ChatGPT, Meta's Llama, Google's Gemini, DeepSeek, Claude, and Grok, providing a comprehensive industry assessment.

Complete Failure in Document and Image Analysis

Grok demonstrated what the ADL termed "complete failure" when tasked with analyzing documents and images containing hateful content, scoring zero in multiple test category combinations. The chatbot's poor performance in multi-turn dialogues reveals it struggles to maintain context and identify bias in extended conversations, severely limiting its utility for customer service and interactive applications.

Broader AI Safety Concerns Across All Models

Grok's performance is particularly alarming, all six models tested demonstrated significant safety gaps. The study underscores the urgent industry-wide need for stronger content moderation safeguards, improved training protocols, and responsible AI development practices to prevent the spread of hateful ideologies.

Rising gas power development due to data center demand
AI data centers’ growing electricity needs are a significant driver behind a recent global increase in gas-fired power plant projects.

U.S. leads global growth in gas capacity

In 2025, gas power capacity under development grew by 31%, with the U.S. surpassing China and making up nearly a quarter of this increase. Over one-third of that U.S. growth is intended to power data centers directly.

Environmental concerns despite lower CO₂ than coal

Although gas emits less carbon dioxide than coal, methane leaks from gas production make it a potent greenhouse gas, complicating climate efforts.

Risk of locking in future emissions

The expansion of gas infrastructure could lock in long-term emissions and create stranded assets if AI energy demand projections don’t materialize.

Policy context and climate goals at stake

The build-out of gas power plants clashes with global climate goals set by agreements like the Paris Accord goals the U.S. has stepped away from under its current administration, which emphasizes fossil fuel expansion.

Start a New Chat

Switch Your AI Model

Try This Prompt & Observe the Output