Safety Index: Anthropic tops rankings as military AI partnerships emer
The Future of Life Institute published its 2026 Summer AI Safety Index on July 1, assessing safety practices across nine major AI companies and surfacing a significant shift in industry policy toward military applications Future of Life Institute.
Safety Rankings and Domain Leaders
Anthropic earned the highest overall grade, leading in five of six assessed domains, while OpenAI took the top position specifically in Risk Assessment Future of Life Institute. Google DeepMind remains a top performer across the index. The rankings reflect measurable differences in how companies document safety practices, conduct risk evaluations, and establish governance frameworks.
Three companies received failing grades in the 2026 assessment: xAI (US), DeepSeek (China), and Mistral (Europe) Future of Life Institute. This marks a notable decline for xAI, which dropped from 4th place in the previous index to 7th place. Meta improved from 6th to 4th place, demonstrating measurable progress in its published safety commitments and governance structures.
Military AI Partnerships as Emerging Risk
Reviewers identified a critical emerging harm risk: the industry's pivot toward military and defense applications. Companies including Anthropic, OpenAI, Google DeepMind, and Meta have reversed previous policies restricting military use cases to actively seek defense partnerships Future of Life Institute. This reversal represents a structural shift in how major labs view their role in the defense sector and signals a departure from earlier commitments to limit weapons applications.
The Index's emergence of military AI as a flagged risk reflects growing scrutiny of how safety assessments account for downstream uses of deployed systems. The assessment documents these partnerships as factual company moves rather than hypothetical risks, establishing a baseline for tracking how safety practices adapt when commercial incentives intersect with national security priorities.
What the Index Measures
The 2026 Summer AI Safety Index evaluates nine companies across six domains of safety governance and practice. The assessment is concrete and measurable—based on documented policies, published commitments, and verifiable safety infrastructure rather than future roadmaps or claimed capabilities. It reflects July 2026 practices, making it a snapshot of industry standards as they stand in mid-2026.
For developers, researchers, and enterprise buyers evaluating AI providers, the Index offers a structured comparison of safety maturity. The inclusion of military partnerships as a tracked emerging risk signals that safety assessment frameworks are evolving to capture business model shifts alongside technical safety measures.
