Key Points
- Anthropic ranked first in the Future of Life Institute's Summer 2026 AI Safety Index with a C+.
- OpenAI and Google DeepMind scored C; Meta a D+; xAI, DeepSeek, and Mistral received failing grades.
- No company earned an A or B across 37 indicators and six safety domains.
- Reviewers said the top labs weakened pledges to pause development at danger thresholds.
The Future of Life Institute released its Summer 2026 AI Safety Index this week, grading nine leading AI companies across 37 indicators and six domains. Anthropic finished first with a C+. No company scored an A or a B.
Behind Anthropic, OpenAI and Google DeepMind each earned a C, Meta a D+, and China's Z.ai and Alibaba Cloud a D-. xAI, DeepSeek, and Mistral received failing grades — one company each from the United States, China, and Europe. Anthropic led five of the six domains, the report noted, on the strength of its transparency, safety framework, and governance work.
The more damning finding sits underneath the letter grades. Reviewers said Anthropic, OpenAI, Google DeepMind, and Meta have all weakened or eliminated earlier commitments to pause development if their systems approached defined danger thresholds — "moving goalpost" behavior, in the panel's words. Over the same 2024-to-2026 window, the same four companies — all of which once banned military applications — reversed course and embraced defense partnerships, a shift reviewers flagged as an emerging harm.
"The capabilities race has become more extreme," said Prof. Stuart Russell, one of the reviewers, per TIME. "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."
The read: a first-place C+ is faint praise. The safest-graded company in the industry still falls short of the bar its own framework once set, and the labs racing hardest are quietly deleting the guardrails they wrote for themselves. The stakes only rise with the money. Anthropic just closed a $65 billion round at a $965 billion valuation, and OpenAI is negotiating everything from a 5% government stake to federal limits on who can use its most capable models. The commercial pressure has never been higher, and the safety scores say the guardrails are moving the wrong way.
Source: Future of Life Institute, TIME.
Comments