When Anthropic CEO Dario Amodei told the Pentagon his company “cannot in good conscience accede” to its contract demands, he drew two explicit red lines: Claude would not be used for mass surveillance of Americans, and it would not power fully autonomous weapons. The Pentagon responded on the record but showed no willingness to revise the contested terms. That standoff, which played out in public view, has since rippled through enterprise procurement offices across the country.
According to spending analyses tracked by industry research firms, Anthropic now holds roughly 25 percent of the business AI subscription market, a share that has grown as OpenAI’s position among large commercial accounts has slipped. The figures, drawn from third-party software spending data segmented by industry vertical, should be read as directional estimates rather than audited results. But the trend they describe is consistent across multiple analyst assessments: compliance-sensitive enterprises are shifting paid AI commitments toward Anthropic, and the company’s refusal to pursue military contracts is a factor procurement teams are raising in vendor evaluations.
Two companies, opposite directions
The divergence between Anthropic and OpenAI on defense work is not subtle. In January 2024, OpenAI quietly updated its usage policy to remove a blanket prohibition on military applications. Within days, reporting confirmed the company was working with the U.S. military on cybersecurity tools for veterans. By late 2024, OpenAI had deepened its defense footprint, partnering with defense technology firm Anduril to integrate AI models into national security systems.
Anthropic moved the other way. Beyond Amodei’s public statement, the company maintains a Responsible Scaling Policy and an acceptable-use framework that restricts how Claude can be deployed. Enterprise customers negotiating contracts with Anthropic encounter these restrictions as binding terms, not just public talking points. For procurement teams at healthcare systems, financial institutions, and universities that operate under heavy federal compliance obligations, that distinction matters.
Why enterprise buyers care
Large organizations do not switch AI vendors over press releases. They switch when a vendor’s policy exposure creates risk that their own compliance teams cannot manage. The concern procurement officers have raised in industry forums and vendor-selection discussions is specific: an AI model that simultaneously serves sensitive defense workloads and commercial customers could attract regulatory scrutiny, congressional attention, or civil-society pressure that spills over onto every company using that model.
That risk calculus has sharpened as AI regulation gains momentum in Washington and Brussels. A vendor entangled in a controversial surveillance or weapons program could become a liability on a compliance audit, even if the enterprise customer’s own use case is entirely benign. Boards and investors increasingly treat these associations as governance risks, and the reputational math has started to influence contract decisions that used to be driven almost entirely by performance benchmarks and pricing.
None of this means ethics is the only variable. Claude’s technical capabilities, its performance on coding and analysis tasks, and Anthropic’s API reliability all factor into enterprise evaluations. OpenAI’s GPT-4o remains the dominant model in many benchmarks, and its ecosystem of integrations gives it structural advantages that do not disappear because of a policy dispute. The shift in subscription share reflects a combination of forces, not a single cause.
What the data can and cannot show
No publicly available enterprise survey has isolated the military-use policy difference as the sole driver of the subscription shift. The market-share estimates that circulate among analysts rely on overlaying third-party spending data onto industry classification codes, a method that introduces assumptions about how software purchases map to specific AI vendors. Different research firms could reach slightly different numbers using the same underlying framework.
OpenAI’s defense portfolio also remains partially opaque. The confirmed scope of its military work covers cybersecurity tools for veterans and the Anduril partnership, but no public Pentagon procurement record has surfaced to clarify contract values, durations, or expansion plans. Whether OpenAI is pursuing additional classified or sensitive programs is not established in available reporting.
On Anthropic’s side, the CEO’s statement is unambiguous, but the company has not released board resolutions or internal policy memos showing exactly how the refusal was communicated to prospective enterprise clients. Enterprise buyers typically require written vendor policies before adjusting procurement strategies, and whether every Anthropic commercial contract now contains explicit military-use exclusions is not confirmed.
The competitive landscape beyond two players
Enterprise AI procurement does not happen in a vacuum occupied only by Anthropic and OpenAI. Google’s Gemini models, available through Google Cloud, carry their own set of AI principles that explicitly prohibit weapons applications while permitting other government work. Microsoft, OpenAI’s largest investor and cloud partner, has long-standing defense contracts through Azure Government and has not signaled any retreat from military AI. Amazon Web Services, which hosts Anthropic’s models through Bedrock, also serves defense and intelligence customers.
For procurement teams evaluating the full vendor landscape, the question is not simply “military or no military.” It is how each vendor defines its boundaries, how those boundaries are enforced contractually, and how exposed a customer becomes if those boundaries shift. Anthropic’s value proposition to compliance-sensitive buyers rests partly on the clarity and specificity of its restrictions. If those restrictions ever soften, the procurement rationale could shift just as quickly.
Where the market goes from here
The split between Anthropic and OpenAI on defense work is documented, accelerating, and now commercially consequential. As of mid-2026, it has moved from the margins of AI ethics debates into the core of enterprise vendor selection. Procurement officers at regulated institutions are asking vendors to put their military-use policies in writing before contracts close, a practice that barely existed 18 months ago.
If Anthropic continues to gain share among compliance-sensitive sectors while holding its red lines, it will strengthen the case that clear boundaries on military and surveillance use can be commercially viable at scale. If the market instead rewards vendors that pursue broad defense portfolios, that outcome will send a different signal about how buyers weigh ethics against performance and price. Either way, the choices these companies have already made are shaping where the next wave of enterprise AI spending lands, and neither side shows any sign of backing down.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
