Anthropic Tops 2026 AI Safety Index, But No AI Firm Earns Above a C+
The report suggests the AI industry's rapid advances continue to outpace progress in safety and governance.
News
- Saudi Payment Disruptions Test UAE–Saudi Trade Ties
- AIRF Dubai 2026 Will Put Agentic AI Execution in the Spotlight
- Microsoft and G42's Inception42 Join Forces on Enterprise Agentic AI
- UK Foreign Secretary Warns World Cannot Wait for 'AI Hiroshima' Before Regulating AI
- UAE Banks Defy Rate Cuts With Double-Digit Profit Growth
- Sam Altman Floats Plan to Give US Govt 5% OpenAI Stake: Report
US artificial intelligence startup Anthropic has eclipsed its competitors to score the highest in a semiannual safety ranking based on public data and information provided by the companies.
The AI Safety Index 2026 by the US-based AI safety think tank Future of Life Institute evaluated efforts across six distinct categories: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing, for Anthropic, OpenAI, Google DeepMind, Meta, Z.ai, Alibaba Cloud, SpaceXAI (formerly xAI), DeepSeek, and Mistral.
Anthropic, the leader in the survey, achieved a grade of C+ with a score of 2.66, while xAI, DeepSeek, and Mistral received an F with scores of 0.65, 0.47, and 0.33, respectively. The Claude-maker led across five of six domains, while OpenAI led Risk Assessment with a score of 2.54.
No company received an “A” in any single category, while Anthropic got the best overall score of “C+.”
Overall, Meta improved its ranking from sixth to fourth, while xAI dropped from fourth to seventh position.
Debuting on the list, Mistral scored last on safety, even though the European Union leads in AI safety regulation. The startup told AFP that the report’s framework isn’t suited for its approach to developing AI models.
The review was conducted by an independent panel of seven researchers and governance experts, comprising David Krueger of the University of Montreal; Sharon Li of the University of Wisconsin-Madison; Tegan Maharaj of HEC Montréal; Sneha Revanur of Encode; Stuart Russell of the University of California at Berkeley; Robert Trager of the University of Oxford; and Yi Zeng of the Gaoling School of AI.
“While there is good work being done on AI safety in the industry, the capabilities race has become more extreme. Companies have backed away from earlier commitments to release new systems only with safety measures appropriate for their capability levels; now, they’re planning to release them even if it’s demonstrably unsafe to do so,” said Prof. Stuart Russell, Professor of Computer Science, UC Berkeley.
As AI models get adopted for military operations, the report flagged the industry’s pivot to military AI use as an emerging current harm risk. Additionally, it noted that while existential safety is the weakest domain industry-wide, constructive attempts exist. While Anthropic’s constitutional classifiers, OpenAI’s call for governance institutions, Google DeepMind’s monitoring commitments, and Meta’s loss-of-control provisions exist, the panelists found them to be “entirely inadequate.”
Three Chinese developers included in the report also produce open models and landed in the bottom half of the ranking: DeepSeek at fifth, Alibaba Cloud at sixth, and Z.ai at eighth.
