No AI Lab Scores Above C- on Existential Safety, Says the Summer 2026 AI Safety Index
TL;DR. The Future of Life Institute’s Summer 2026 AI Safety Index graded nine AI companies. Anthropic had the highest overall grade, a C+ (2.66). On existential safety, the Index says it is “the weakest domain industry-wide” and “No company exceeds C-; most score D or below.”
The overall grades
The page is headed July 2026, with evidence collected through 3 June 2026.
| Company | Overall grade | Score |
|---|---|---|
| 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 |
| Mistral | F | 0.33 |
What the Index concludes
- Existential safety is “the weakest domain industry-wide”: “No company exceeds C-; most score D or below.”
- Anthropic again has the top overall grade. OpenAI now leads the Risk Assessment domain.
- Meta moved from 6th to 4th place, and xAI from 4th to 7th.
- The three failing grades are one each from the US (xAI), China (DeepSeek) and Europe (Mistral).
- Anthropic, OpenAI, Google DeepMind and Meta have, in the Index’s assessment, weakened or voided pledges to pause unilaterally.
- The Index also notes an industry-wide pivot to military uses of AI.
- On interpretability and chain-of-thought monitorability the Index’s criticism is that “Detection is not prevention.”
The fine print
- These are one organisation’s grades, produced by its own rubric and reviewers. Another rubric could rank the companies differently.
- We report overall grades and quoted conclusions only. Domain-by-company grades sit in a scorecard we did not map line by line, so we leave them out.
- Our reading: a low grade on a particular rubric is not a direct measure of how likely harm is.
- Evidence was collected through 3 June 2026; the situation at any company may have changed since.
Sources
What we read: The Index web page (the grade summary and key conclusions, verbatim). We did not open the full PDF.
Scenda.AI comments on research by others. We are not the authors of the papers and reports covered here, and nothing on this page is endorsed by them.