The Future of Life Institute (FLI), a U.S.-based nonprofit focused on reducing large-scale technological risks, published its Summer 2026 AI Safety Index on July 7, 2026, delivering a broadly critical verdict on the safety practices of nine major artificial intelligence developers worldwide.
Methodology and Scope
<cite index="11-14,11-15,11-16">The index evaluated nine companies — Anthropic, OpenAI, Google DeepMind, xAI, Z.ai, Meta, DeepSeek, Alibaba Cloud, and Mistral — using evidence collected up to June 3, 2026, drawn from publicly available materials including model cards, research papers, and benchmark results, supplemented by a targeted company survey. An independent panel of seven leading AI researchers and governance experts assigned domain-level grades on an A-to-F scale based on absolute performance standards.</cite> <cite index="18-12">The six evaluated domains were risk assessment, current harms, safety frameworks, existential safety, governance, and information sharing.</cite>
Grades and Rankings
<cite index="14-1,14-3">Anthropic earned the top overall grade of C+ with a numerical score of 2.66, while xAI, DeepSeek, and Mistral received failing grades of F, with scores of 0.65, 0.47, and 0.33, respectively.</cite> <cite index="15-10,15-11">OpenAI and Google DeepMind each received a C, while Meta managed only a D+. Z.ai and Alibaba Cloud scored a D-, and xAI, DeepSeek, and Mistral received the lowest possible grade of F. Not a single company earned an A or B.</cite>
<cite index="6-18,6-19">Anthropic earned the best overall grade and led on risk assessments, including conducting human-participant bio-risk trials, excelling in privacy practices, carrying out alignment research, and demonstrating governance commitment through its Public Benefit Corporation structure.</cite> <cite index="11-5">OpenAI led specifically in risk assessment on the strength of a broader evaluation suite and diverse engagement with external testing.</cite>
<cite index="11-6">Meta improved from sixth to fourth place overall, while xAI dropped from fourth to seventh.</cite> <cite index="17-12,17-13">The largest single-cycle decline came from xAI, which fell to a failing grade after merging into SpaceX; reviewers noted the merger made it harder to track safety practices and that the company still lacks basic transparency and governance structures.</cite>
Systemic Weaknesses
<cite index="19-4,19-5">The index found that leading developers are walking back earlier safety commitments: Anthropic, OpenAI, Google DeepMind, and Meta have all weakened or voided pledges to pause development unilaterally if certain risk thresholds were crossed, with some pauses now made contingent on competitors' actions.</cite>
<cite index="19-21,19-22,19-23">Existential safety — covering long-term risks from highly capable or self-improving systems — was the industry's weakest domain overall. No company scored above a C-, and most fell to a D or lower. The report credited constructive attempts, including Anthropic's constitutional classifiers, OpenAI's calls for new governance institutions, Google DeepMind's monitoring commitments, and Meta's provisions against loss of control, but judged them "entirely inadequate."</cite>
<cite index="18-6">The index highlights that amid intense market competition, progress on commercial safety has largely plateaued or regressed, with multiple labs scaling back prior safety commitments to pursue national security and defense contracts.</cite>
Military Use and Current Harms
<cite index="19-13,19-14,19-15">A related finding concerned the industry's shift toward military applications. Companies including Anthropic, OpenAI, Google DeepMind, and Meta had previously restricted or banned military use of their systems but reversed course between 2024 and 2026. The panel singled out Anthropic for "questionable military engagements," citing a reported link to the Minab school strike, which caused mass civilian deaths.</cite>
Governance and Regulatory Context
<cite index="19-8,19-9,19-10">FLI safety investigator Sabina Nong noted that competitive pressure to release increasingly capable models has intensified, with Anthropic and OpenAI now tying deployment pauses to competitors' actions, and Google DeepMind and Meta having removed unilateral pause commitments altogether. "If we leave AI safety to corporate-defined rules, we are destined to race to the bottom," she said.</cite>
<cite index="2-18,2-19,2-20">The broader pattern the index documents is that AI capabilities are improving far faster than safety. FLI president Max Tegmark has drawn an analogy to clinical trials or nuclear safety — domains where regulators demand quantitative evidence and rigorous controls before deployment — arguing that AI, arguably more powerful and more general, has no equivalent framework.</cite>
The Summer 2026 index is the fourth edition FLI has published since launching the series in 2024, and the first to include European developer Mistral, which finished last among all nine companies evaluated.