Full Breakdown
Global AI Safety Index Exposes Industry-Wide Shortfalls
7/8/2026, 6:38:48 AM
AI Safety Index Release and Prior Pledges
On July 7, the Future of Life Institute (FLI) released its “AI Safety Index for the First Half of 2026,” evaluating nine leading AI developers—Anthropic, OpenAI, Google DeepMind, Meta, Z.ai, Alibaba Cloud, xAI, DeepSeek and France’s Mistral. The assessment follows earlier “red-line” pledges to halt military-related AI work, which FLI says have softened as several firms gradually reversed bans.
Rankings and Grades
The index assigned grades from A to F. Anthropic led with a C+, OpenAI and Google DeepMind each earned a C, Meta received D+, Z.ai and Alibaba Cloud D-, while xAI, DeepSeek and Mistral each scored an F. No company achieved an A or B, indicating uniformly low safety performance across the sector.
Official Findings and Company Responses
FLI concluded that the safety measures of even the highest-ranked firms fall far short of the rapid pace of AI advancement. The report noted Anthropic’s February withdrawal of a pledge not to train systems without proven safety and cited questionable military engagements. Mistral argued that the index’s framework does not suit its open-model approach. Alibaba Cloud, xAI and DeepSeek did not respond, and Anthropic declined further comment.
Criticism and Conflicting Data
Mistral’s objection underscores concerns that the ranking methodology may not capture nuances of open-model development. The lack of responses from five firms limits data completeness. Moreover, Anthropic, despite topping the index, is reported to have supplied technology to U.S. military operations in Venezuela and Iran after a Pentagon ban was lifted on June 30. These contradictions illustrate gaps between safety scores and real-world deployments.
Implications for Security and Regulation
The index warns that AI risks have shifted from speculative to present-day security challenges, including the potential for artificial general intelligence and immediate harms such as cyber-attacks. The findings suggest voluntary industry pledges are insufficient, reinforcing calls for external oversight, stronger governance mechanisms, and policy measures to curb unchecked development.
Verbatim Quotes
- “I was very disappointed to find that they came last, especially since Europe has really... been a leader in AI safety,” — Max Tegmark, MIT professor and FLI president
- “We reached out many, many times” — Max Tegmark, FLI president
- “Mistral was included on the list for the first time, though when asked by AFP to comment on its last place, the company said the report’s framework is not suited for its approach to developing AI models.” — Mistral spokesperson
- “companies are backtracking on the 'red line' pledges they made in the past.” — Future of Life Institute
Outlook: Next Steps and Policy Considerations
FLI urges reversal of weakened safety measures and advocates for regulatory frameworks that enforce compliance beyond voluntary pledges. Ongoing monitoring of military applications and the development of artificial general intelligence are identified as priority areas for future investigation.
