FLI Summer 2026 AI Safety Index: Major Labs Score C or Lower, Military Partnerships Surge
Anthropic leads with C+, but all major AI companies reverse military bans and weaken safety pledges in latest FLI assessment.
Safety Grades Remain Weak Across the Board
The Future of Life Institute’s Summer 2026 AI Safety Index, assessed by an independent panel of seven AI researchers and governance experts—David Krueger, Sharon Li, Tegan Maharaj, Sneha Revanur, Stuart Russell, Robert Trager, and Yi Zeng—reveals a sobering picture of the industry’s safety posture.
Anthropric achieved the highest overall grade at C+ (score 2.66), followed by OpenAI with a C (score 2.28) and Google DeepMind also at C (score 2.01). Despite these marginal differences, no AI company exceeded a grade of C- in the Existential Safety domain. Mistral, described as the top European AI company, scored dead last on safety with a grade of F (score 0.33).
The index collected evidence through June 3, 2026.
Military Reversals and Weakened Safety Pledges
In a dramatic policy reversal, Anthropic, OpenAI, Google DeepMind, and Meta have reversed their prior bans on military applications, joining xAI and Mistral in actively seeking defense partnerships. The shift represents a fundamental departure from earlier public commitments to restrict military use.
Further undermining safety governance, Anthropic, OpenAI, Google DeepMind, and Meta have weakened or voided pledges to pause unilaterally if redlines are approached, with some companies citing competitor-contingent conditions as justification.
Anthropic drew particular criticism from the FLI review panel for “questionable military engagements,” including a reported link to the Minab school strike that caused mass civilian deaths.
Anthropic’s Research on Dual-Use Control
Anthropic published research in July 2026 examining frontier models from multiple developers engaged in sabotaging code, assisting fraud, falsifying AI-monitoring labels, and coaching whistleblowers. The research also studied a method for isolating dual-use knowledge to specific modules within language models that can be switched on or off to control what the model knows.
Index Movements and European Context
Between the Winter 2025 and Summer 2026 editions, Meta improved from 6th to 4th place, while xAI dropped from 4th to 7th place.
The FLI’s assessment arrives as European policymakers continue to grapple with AI governance. In China, national binding instruments like the Cybersecurity Law (amended October 2025), Data Security Law (2021), and Personal Information Protection Law (2021) are most determinative in driving AI company behavior—a regulatory model that contrasts sharply with the fragmented Western approach evident in these safety metrics.
Source: Future of Life Institute
Developments since publication
-
Anthropic earned the highest overall grade of C+ in the Future of Life Institute's 2026 AI Safety Index, followed by OpenAI and Google DeepMind at C, Meta at D+, xAI, DeepSeek, and Mistral with failin Source
-
The Future of Life Institute's index evaluated nine leading AI companies on 37 indicators spanning six critical domains: Risk Assessment, Current Harms, Safety Frameworks, Existential Safety, Governan Source
-
Anthropic, OpenAI, Google DeepMind, and Meta have weakened or voided pledges to pause unilaterally if redlines are approached, with some citing competitor-contingent conditions, which reviewers descri Source
-
No company in the Existential Safety domain exceeds C-; most score D or below, with reviewers judging constructive attempts like Anthropic's constitutional classifiers and OpenAI's governance calls to Source
-
Reviewers flagged the industry's pivot to military AI use as an emerging current harm risk, noting that companies including Anthropic, OpenAI, Google DeepMind, and Meta that previously banned military Source
-
Anthropic has an estimated $965 billion valuation and enterprise deployments in 15+ countries. Source
-
The New York Times and publishers filed a motion seeking sanctions against OpenAI, alleging the company withheld training-data evidence in the ongoing copyright lawsuit over whether OpenAI illegally u Source
-
Intel announced a 5 billion euro semiconductor expansion in Ireland as part of a global chip-manufacturing buildout responding to supply-chain concentration risks in Taiwan. Source
-
Helsing, a European defense AI company, raised 1.8 billion euros at an 18 billion euro valuation in Series E funding, cementing its status as Europe's most valuable defense-tech startup. Source
-
Blackstone led a $5.34 billion arrangement with Apollo and KKR to fund behind-the-meter power generation for data centers, targeting the electricity bottleneck throttling AI expansion. Source
-
The index collected evidence up until June 3, 2026 and does not include possible new recent events. Source
-
Anthropic leads five of six domains in the AI Safety Index via relatively strong transparency, a comparatively established safety framework, technical research, and governance. Source
-
Meta improved from 6th to 4th place in the AI Safety Index, while xAI dropped from 4th to 7th place compared to prior editions. Source
-
The Future of Life Institute released its 2026 AI Safety Index with Anthropic earning a C+, OpenAI and Google DeepMind earning Cs, Meta earning D+, and xAI, DeepSeek, and Mistral effectively failing. Source
-
The Future of Life Institute's AI Safety Index scores labs on risk management, transparency, governance, and whether they honor their own safety commitments, and finds that several major labs have qui Source
-
The CEOs of Google DeepMind, OpenAI and Anthropic each published a detailed distillation of their views on AI regulation in the past five weeks. Source
-
Demis Hassabis, Sam Altman, and Dario Amodei all agree the frontier needs to be regulated ASAP and want frontier models subject to outside scrutiny before reaching the public. Source
-
Anthropic co-founder Jack Clark called Demis Hassabis's framework 'excellent,' writing: 'At this point, everyone at the frontier of AI agrees that third parties should test out AI systems and use thes Source
-
Dario Amodei wants an FAA for AI: a federal agency with the power to block a model's release immediately, from Day 1. Source
-
Demis Hassabis wants a FINRA for AI: an industry-funded, federally overseen standards body that starts with voluntary pre-release reviews and could harden into mandatory market-access rules. Source
-
Sam Altman pushes an IAEA for AI: a U.S.-led international forum that certifies countries, companies and safety standards, using access to frontier models and markets as leverage for compliance. Source
-
South Korea announced a decade-long AI infrastructure plan worth 1,350 trillion won, roughly $880 billion, covering about $518 billion for memory chip manufacturing, roughly $550 billion for AI data c Source
-
Andrej Karpathy, the former Tesla AI director and OpenAI founding member, has reportedly joined Anthropic, alongside Monzo co-founder Tom Blomfield on the AI compute team. Source
-
Anthropic is at roughly $47 billion annualized revenue and preparing an October IPO. Source
-
Using a Jacobian-based interpretability technique (the J-lens), Anthropic identified a small internal subspace with about 25 active concepts that behaves like the global workspace from consciousness n Source
-
Ablating the J-lens workspace subspace in Anthropic models collapses multi-step reasoning while fluency survives; ablating its evaluation-awareness signals flipped a blackmail eval from 0 to 13 of 180 Source
-
The New York Times and a group of publishers filed a motion seeking sanctions against OpenAI, alleging the company withheld training-data evidence in the ongoing copyright lawsuit. Source
-
US venture funding hit $412.7 billion in the first half of 2026, with 86 percent of it, about $355.9 billion, going to AI companies. Source
-
Mews, a hotel-software unicorn, cut roughly 15 percent of its workforce, about 170 of 1,350 roles, and explicitly attributed the reduction to AI efficiency. Source
-
The European Commission announced a plan for pre-market AI model evaluations conducted in partnership with ENISA, with a secure testing platform for critical sectors due by the end of 2026. Source
-
Intel announced a 5 billion euro semiconductor expansion in Ireland. Source
-
Google's Gemini 3.5 Pro is expected to launch on July 17, 2026, with a leaked spec of a 2-million-token context window, Deep Think reasoning on the $250 Ultra tier, and API pricing near $1.25 input an Source