Imagine a classroom where the smartest kid gets a C+. That's the Summer 2026 AI Safety Index. Nine of the most powerful AI companies on Earth — OpenAI, Anthropic, Google DeepMind, xAI, Meta, DeepSeek, Mistral, and others — were graded across 37 indicators. The highest mark? A C+ for Anthropic. The rest? C's, D's, and outright fails. This isn't a report card. It's a warning siren.
The Index, from the Future of Life Institute, measures visible safety practices, not raw model capability. So a company can score well on paper — policies, frameworks, disclosure — while shipping riskier models. That's like getting an A in driver's ed because you memorized the manual, then street racing on the way home. Anthropic's 2.66 out of 4 might look like a win, but it's a win in a race where everyone else is limping. OpenAI and Google DeepMind tied for second with a C, but their scores (2.28 and 2.01) differ — proof that letter grades compress real gaps. Meanwhile, xAI, DeepSeek, and Mistral reportedly failed outright. That's Elon Musk's outfit lumped with China's DeepSeek and France's Mistral. Awkward.
But here's the twist: the Index grades transparency, not actual safety. A lab with a safer model but poor disclosure could rank lower than a riskier lab that documents everything. So the leaderboard rewards paperwork as much as protection. That's a problem when capabilities and defense use are expanding faster than any safety ceiling. As one report put it, the sharper point isn't who leads — it's how weak the ceiling remains. And with pause pledges eroding, the whole class is sliding.
Why should you care? Because these companies are building the AI that will soon drive your car, diagnose your illness, and fight wars. If their safety practices are barely passing, we're all in the backseat of a driverless car with a C+ student at the wheel. The Index could become a regulatory cudgel in Brussels and Washington. Failing grades might cost procurement deals and enterprise trust. Anthropic's C+ might win it government contracts, but it also inherits the burden of proof — being the best of a bad bunch isn't exactly a victory lap.
What to watch? The Fall/Winter 2026 edition. If Anthropic slips below C+, that's a signal that commercial pressure is crushing safety pledges. If someone breaks into B-territory, it might validate that transparency pressure works. And if governments keep buying from D+ and F-rated labs, the Index becomes a paper tiger. The industry graded its own homework — and still barely passed. That should terrify you. The question isn't who won. It's whether anyone can do better before it's too late.