Anthropic tops AI safety index with C+, OpenAI close behind with a C

The AI industry’s best safety performers just earned the equivalent of a “meets minimum expectations” on their report card. Anthropic and OpenAI claimed the top two spots in the Future of Life Institute’s Summer 2026 AI Safety Index, but their grades, C+ and C respectively, suggest the bar for “best in class” remains remarkably low.

Anthropic led the pack with a score of 2.66 out of 4.0, while OpenAI trailed slightly at 2.28.

The full scorecard

The FLI index evaluates leading AI developers across six domains and 37 specific indicators, covering everything from transparency and governance to risk assessment and safety frameworks.

Google DeepMind earned a C with a score of 2.01, placing it in third position. Meta scored a D+ at 1.67. xAI, DeepSeek, and Mistral all received outright F grades.

Anthropic distinguished itself by leading in five of the six evaluation domains, with particularly strong marks for transparency, governance, and its Responsible Scaling Policy. OpenAI’s best showing came in the Risk Assessment domain, where it outperformed all competitors.

The index is produced biannually by a panel of experts that includes Stuart Russell and David Krueger. It uses a US GPA-style grading system.

Safety pledges are going in the wrong direction

Perhaps more troubling than the mediocre grades is a trend the report identified: safety commitments among the top-ranked firms are actually declining. The companies that should be setting the standard are quietly walking back previous pledges.

The report also flagged an increase in military AI collaborations across the industry.

What the grades mean for the AI landscape

The fact that no company managed to crack a B-minus tells us something fundamental about the state of AI safety. The entire industry is clustered in mediocrity or worse, with F-grade labs shipping products alongside C+ labs competing for the same markets and talent pools.

Anthropic’s Responsible Scaling Policy, which outlines specific capability thresholds that trigger additional safety measures, appears to have contributed meaningfully to its leading score.

DeepSeek in particular has gained significant market traction, which means its safety practices, or lack thereof, affect a growing user base.

The FLI’s next index will arrive in early 2027.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.