The AI Safety Index 2026 is out — and the results should make everyone stop scrolling. When the world’s leading AI watchdog handed out grades this week, the best company in the entire industry barely scraped a C+. Every other lab did worse, and three of them failed outright.

Released by the Future of Life Institute, the Summer 2026 edition of the AI Safety Index graded the nine biggest AI labs on Earth across 37 safety indicators. Not a single one scored above a C+. Here’s what the report card really means.

What the AI Safety Index 2026 Actually Measures

The AI Safety Index 2026 judged Anthropic, OpenAI, Google DeepMind, Meta, xAI, DeepSeek, Mistral, Alibaba Cloud, and Z.ai across six domains — risk management, transparency, governance, and whether each company honors its own safety commitments. In total, 37 indicators went into every grade.

The Grades: Nobody Passed With Honors

• Anthropic — C+ (2.66 / 4.0). The top of the class — and it’s still a C+.
• OpenAI — C (2.28)
• Google DeepMind — C (2.01)
• Meta — D+
• xAI, DeepSeek, and Mistral — failing grades

One failure each from the US, China, and Europe. Cutting corners on AI safety, it turns out, is a truly global phenomenon.

The Most Alarming Finding in the AI Safety Index 2026

The grades are bad. But the scariest part of the AI Safety Index 2026 is what the top companies have stopped promising to do.

Labs had previously committed to release new AI systems only with safety measures appropriate to how capable those systems are. According to the index, several are now walking that back — planning to ship powerful models even when it may be demonstrably unsafe to do so.

In other words, the people building superhuman AI are, by the assessment of independent researchers, backing away from their own safety promises — in the middle of the most competitive race the tech industry has ever seen.

Why AI Safety Keeps Losing

The reason is simple: speed sells. With rivals shipping frontier models every few weeks and billions in market value riding on each launch, “let’s slow down to be safe” is a losing move on an earnings call. Safety commitments made in calmer times are the first thing renegotiated when the pressure is on.

What It Means for You

We are handing more of our lives — our code, our medicine, our infrastructure, our information — to systems built by companies a watchdog just graded between “mediocre” and “failing.” The AI Safety Index 2026 is a reminder that the technology is sprinting ahead while the safety walks backwards.

Does a C+ make you trust these companies less — or is this just the price of progress? Let us know in the comments.