Future of Life Institute Grades 9 AI Labs on Safety: The Highest Score Is a C+

My take: The Future of Life Institute evaluated nine major AI laboratories across six safety dimensions, and not one earned an A or a B. The highest score was a C+, earned by Anthropic. OpenAI and Google DeepMind received a C, Meta managed a D+, and xAI, DeepSeek, and Mistral received failing grades. The institute is direct about it: "even the top-ranked companies' current safety measures are completely inadequate relative to the pace of AI capability advancement."

What I think is worth underscoring is that this is not an accusation of dishonesty. It is a diagnosis of speeds: model capabilities are advancing faster than the safety frameworks being built to contain them. And that is precisely the kind of risk you cannot evaluate from the outside if you rely only on what each company says about itself.

For any company or organization deploying AI in production today, this is a clear signal: you cannot delegate all risk management to your vendor. Internal controls matter — knowing which tools you use, what data you feed them, and how you evaluate their outputs before they reach your customers or critical decisions.

Does your team have its own process for evaluating the risk of the AI models you use, beyond what each vendor publicly declares?