The Future of Life Institute released its Summer 2026 AI Safety Index on July 7 — and the headline number says everything: not one of nine leading AI companies scored above a C+. Anthropic topped the ranking with that C+, OpenAI and Google DeepMind each scored a C, Meta a D+, and xAI, DeepSeek, and Mistral all received failing grades. The report lands the same week as Shanghai’s World AI Conference, where China is making its own AI-leadership statement — together, the two stories capture where the industry actually stands going into the back half of 2026: capability is sprinting ahead of safety, and no one, anywhere, is close to an A.
How the grades broke down
An independent panel of seven AI safety and governance researchers — affiliated with institutions including UC Berkeley, Oxford, and the University of Montreal — scored nine companies across 37 indicators in six domains: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing. The final scores:
- Anthropic — C+ (2.66), leading five of six domains on transparency, safety framework maturity, and governance
- OpenAI — C, leading Risk Assessment specifically on the strength of its evaluation suite
- Google DeepMind — C
- Meta — D+, improved from 6th to 4th place overall
- Z.ai and Alibaba Cloud — D-
- xAI — F (0.65), dropped from 4th to 7th place
- DeepSeek — F (0.47)
- Mistral — F (0.33), the lowest score despite being based in the EU, currently the most AI-safety-regulated jurisdiction in the world
That last point is worth sitting with: Mistral scoring dead last while operating under Europe’s AI Act shows that regulatory environment alone doesn’t determine a company’s safety practices — inadequate safety is a global pattern, not a regional one.
The two flagged trends
Reviewers called out two concerning shifts since the last index. First, several companies — including Anthropic, OpenAI, Google DeepMind, and Meta — have gradually walked back previous bans on military AI applications, joining xAI and Mistral in actively pursuing defense contracts. Second, existential safety remains the industry’s weakest domain across the board: the panel described current mitigation efforts, including Anthropic’s constitutional classifiers and OpenAI’s calls for governance institutions, as “entirely inadequate” relative to the risks they’re meant to address.
The same week: WAIC and the multi-superpower AI race
The safety report’s release coincides with Shanghai’s 2026 World AI Conference, where President Xi Jinping is attending in person for the first time since the event began in 2018 — a signal that Beijing is treating AI leadership as a top-tier national priority. It follows South Korea’s separate commitment of $880 billion to AI development over the next decade, and comes in the same stretch as reports that Andrej Karpathy has joined Anthropic and that the New York Times has asked a judge to sanction OpenAI in its ongoing copyright litigation. Taken together with a safety index where the industry leader only manages a C+, the throughline for mid-2026 is a familiar one: the race for capability is intensifying on every front, while the safety consensus that’s supposed to govern it keeps eroding.
Why this matters if you’re choosing which AI tools to use
None of this changes which model is fastest or cheapest for your workflow — but if governance, data handling, or vendor risk factor into your tool decisions, this is a useful independent reference point rather than marketing copy from any one lab. Anthropic’s lead on transparency and safety framework maturity lines up with how it’s generally positioned itself competitively; xAI, DeepSeek, and Mistral’s failing grades are worth knowing if you’re evaluating those providers for anything beyond casual use.
Last updated: July 17, 2026.