OpenAI Signs Pentagon AI Deal Anthropic Refused

A Fault Line in the AI Industry

OpenAI announced on February 28 that it had reached a deal allowing the U.S. military to use its artificial intelligence technologies in classified operations. The agreement came after weeks of tense negotiations that saw Anthropic — OpenAI's chief rival and the maker of Claude — refuse the Pentagon's terms and walk away from the table entirely.

The diverging paths of the two companies represent the most consequential split in the AI industry's relationship with military power. Where OpenAI found compromise, Anthropic found red lines it would not cross, setting the stage for a confrontation that extends well beyond business strategy into fundamental questions about how artificial intelligence should be governed.

What Anthropic Refused

The core dispute centered on the Pentagon's desire to use AI systems for analyzing bulk data collected from American citizens and for applications that could lead to lethal autonomous weapons. Anthropic maintained that these uses violated its core principles and posed unacceptable risks.

Reports indicate the Pentagon wanted broad latitude to deploy Claude for essentially any legal military purpose, a scope that Anthropic viewed as dangerously vague. The company's leadership argued that permitting mass surveillance applications and refusing to draw firm lines around autonomous lethal systems would undermine the safety commitments that define Anthropic's identity.

Negotiations between Anthropic and the Department of Defense reportedly made virtually no progress on these fundamental issues. The gap between what the Pentagon demanded and what Anthropic would accept proved unbridgeable, with neither side willing to compromise on what each viewed as non-negotiable principles.

Innovation

A Nature study in mice suggests low-protein diets can trigger gut microbe-driven signals that turn white fat toward a calorie-burning beige state, revealing a complex microbiome link in metabolism.

DT Editorial AI·Apr 17, 2026·via refractor.io

Innovation

Vertical Aerospace says its full-scale tilt-rotor aircraft completed a continuous two-way transition between vertical and wingborne flight under UK civil aviation oversight, marking a major certification milestone.

DT Editorial AI·Apr 17, 2026·via newatlas.com

Innovation

A new study found scant evidence for endogenous DMT in rat brains and no sign that serotonin terminals formed or retained the compound, reopening a contentious scientific argument rather than settling it.

DT Editorial AI·Apr 17, 2026·via refractor.io

Innovation

Anthropic’s latest model, Mythos, is being made available only to a limited group of organizations after internal and external testing suggested it can autonomously exploit vulnerabilities at a higher level than previous

The Broader Stakes

The OpenAI-Anthropic split on military AI use illuminates a tension that has simmered since the earliest days of the modern AI boom. Technology companies developing the most powerful AI systems in history must decide what constraints, if any, they will place on how those systems are used — and by whom.

This decision carries weight because the capabilities of large language models extend far beyond simple text generation. These systems can analyze vast datasets, identify patterns in intelligence reports, assist in targeting decisions, and potentially operate with decreasing human oversight. The question of who controls these capabilities, and under what rules, is arguably the most important governance challenge the AI industry faces.

The Pentagon's position is straightforward: national security demands access to the best available technology, and companies that develop transformative AI capabilities have an obligation — or at least a strong incentive — to support defense applications. Military leaders argue that adversaries including China are integrating AI into their military systems without ethical hand-wringing, and that American restraint amounts to unilateral disarmament.