The 2026 AI Safety Index Is In: Nobody Scored Above a C+

What the Index measured

The Future of Life Institute's Summer 2026 AI Safety Index is a report card, and this year the whole class is failing to impress. It evaluates nine leading AI companies — Anthropic, OpenAI, Google DeepMind, xAI, Meta, DeepSeek, Mistral, Z.ai, and Alibaba Cloud — across 37 indicators grouped into six domains: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing.

It is graded by people, not a vendor survey. The review panel was seven independent experts led by Stuart Russell of UC Berkeley, joined by David Krueger, Sharon Li, Tegan Maharaj, Sneha Revanur, Robert Trager, and Yi Zeng. That independence is the point: it is an outside read on companies that mostly grade their own safety work.

The grades, in order

Per the Institute's published scores, the ranking runs: Anthropic C+ (2.66), OpenAI C (2.28), Google DeepMind C (2.01), Meta D+ (1.32), Z.ai D- (0.88), Alibaba Cloud D- (0.87), xAI F (0.65), DeepSeek F (0.47), and Mistral F (0.33).

Two facts stand out. First, no company earned an A or a B — the ceiling for the entire frontier industry this cycle was a C+. Second, Anthropic did not just win on the total; it led five of the six domains, on the strength of comparatively established transparency, safety frameworks, technical research, and governance. As IEEE Spectrum noted in its coverage, the bottom of the table — three outright F grades — is as notable as the top.

Why the top grade is still a warning

The headline is not that Anthropic won. It is that winning meant a C+. The Institute's own conclusion is that even the top-ranked companies' current safety measures are inadequate relative to how fast capabilities are advancing.

Russell put the core concern bluntly: companies, he said, have backed away from earlier commitments to release new systems only with safety measures appropriate for their capability levels, and are now planning to release them even when it is demonstrably unsafe to do so. TIME framed the same finding as a report card where nobody gets an A. The Index is not measuring whether these systems are useful — plainly they are — but whether the safety practices are keeping pace, and the panel's answer is no.

What a working reader should take from it

If you buy, deploy, or build on these models, the Index is a diligence input, not a verdict. It grades company safety practices — transparency, frameworks, governance — not the raw capability or price of any given model. A C+ safety grade and a strong coding benchmark can live in the same company at the same time.

The useful move is to read the domain breakdown for the vendor you actually use and see where it is thin. A firm that scores well on governance but poorly on information sharing tells you something concrete about what to ask for in a contract or a model card. Treat the grades as a map of where to push during procurement, not as a reason to switch models on their own.

Sources: Future of Life Institute — AI Safety Index, Summer 2026; IEEE Spectrum — coverage of the 2026 safety grades; TIME — Nobody gets an A on AI safety. Analysis and framing by Precision AI Academy.

Common questions

Who publishes the AI Safety Index? The Future of Life Institute, a nonprofit. The Summer 2026 edition was graded by a panel of seven independent experts led by Stuart Russell of UC Berkeley.

Which company ranked highest? Anthropic, with an overall C+ (2.66), leading five of the six domains. OpenAI (C) and Google DeepMind (C) came next. No company scored an A or a B.

Which companies scored an F? xAI (0.65), DeepSeek (0.47), and Mistral (0.33) each received an overall F grade in the Summer 2026 Index.