AI industry fails on existential safety, index warns
TL;DR:
- The Future of Life Institute’s semiannual safety index ranked nine leading AI companies; none scored above a C+.
- Anthropic topped the ranking but was criticised over “questionable military engagements”.
- Every firm was judged “entirely inadequate” at managing existential risks such as advanced AGI.
No major AI developer is doing enough to contain the technology’s most serious risks, according to the latest Future of Life Institute safety ranking. The US-based think tank scored nine leading companies across six categories, from current harms to existential safety, and awarded no firm higher than a C+, handed to top-placed Anthropic.
A whole industry marked down
The report’s central charge is uniform failure at the top end of risk. All nine companies, it found, are falling short on “existential” threats such as pursuing human-level artificial general intelligence, with efforts “entirely inadequate” even where “constructive attempts exist”. Seven researchers and governance experts built the rankings from public data and company submissions, though Alibaba, xAI, DeepSeek and France’s Mistral did not respond to the survey; Mistral, placed last, said the framework did not suit its open-model approach. The index also flagged a drift back toward military work, criticising Anthropic for “questionable military engagements” after its technology was reportedly used in US operations, even as the Pentagon separately banned the company over safety disagreements.
Looking forward
The findings land squarely in a live UK debate. Britain has positioned its AI Security Institute as a global authority on frontier-model evaluation, and the index’s verdict, that self-governance is not working, strengthens the case for independent testing of the kind UN-led governance efforts are now pressing for. It also arrives days before OpenAI’s GPT-5.6 launch, a reminder that capability continues to outrun the safeguards meant to contain it. For UK firms weighing which models to build on, a public scorecard on safety, however contested, is becoming another factor in the decision.