Imagine getting a report card back. Top grade in the whole class: a C+.
That's roughly where the big AI companies stand right now. The Future of Life Institute — an independent nonprofit that has published a twice-yearly AI Safety Index since 2024 — just released its Summer 2026 report. Nine companies, six grading categories, one conclusion: nobody earned an A. Not even a B.
Who scored what
The ranking, translated into plain grades:
- Anthropic (Claude): C+ — top of the class, with the most transparent processes and the most developed safety framework
- OpenAI (ChatGPT) and Google DeepMind (Gemini): both C
- Meta (Meta AI): D+ — still a jump from 6th to 4th place compared to the last round
- Z.ai and Alibaba Cloud: both D− — passing, but on academic probation
- xAI (Grok): F — dropped from 4th to 7th place
- DeepSeek and Mistral (Le Chat): also F
Three out of nine companies are failing, and two more are hanging on by a thread with their D−. And when the best student in the class only manages a C+, a real school would haul the whole class in for a stern parent-teacher meeting.
The real problem: existential safety
One category scores worst across the entire industry: how companies handle the big, long-term risks — systems that improve themselves, or loss of control over highly capable models. No company scores above a D- here. Most land at F or close to it.
Sounds like science fiction? Maybe. But that's exactly the point of the report: these companies are building systems whose risks they can't properly assess themselves yet — and still treat the topic as a lower priority.
Second red flag: several companies that previously ruled out military use of their models — including Anthropic, OpenAI, Google DeepMind, and Meta — have loosened or dropped that rule since 2024. You don't need to turn this into a scandal. But it's a shift worth noticing if you take these companies' safety promises at face value.
What does this mean for you as a user?
First: no need to panic. An F on the safety index doesn't mean a product is dangerous for you personally as a chat user — the categories mostly measure governance, research transparency, and how future risks are handled, not how the tool behaves in your everyday conversation.
Still, the ranking is a useful compass if you're already choosing between tools: Claude, ChatGPT, and Gemini are clearly ahead on transparency and safety processes. That's not a free pass for blind trust — but it's a difference worth knowing before you type sensitive data into any chat window.
Interesting side note: being European doesn't automatically shield you from an F. Mistral, the French ChatGPT competitor behind Le Chat, sits in the same failing category as xAI and DeepSeek. GDPR compliance for data processing is a different issue from safety research and governance — don't confuse the two just because both get filed under "trustworthy."
And just like with the BigTech privacy comparison: proportion beats alarmism. None of these companies is the final boss, and none is an angel. It's a scale — and right now, the whole industry sits somewhere around "meh."
For all the criticism, one fair question remains: is the bar even set fairly for such a young industry? A good share of these providers are fledgling start-ups, none of which currently turns a profit (we broke down who actually makes money off the AI boom here). A company still fighting for customers and its own survival pours its millions into the product first — not into a department for risks that might only bite ten years from now. That doesn't excuse sloppy safety work. But it partly explains it — and it makes the lead of the few who take it seriously anyway all the more remarkable.
The full report is free from the Future of Life Institute — worth a look if you want more than just the grades. And next time an AI company proudly claims "safety is our top priority": check the "report card" first.