US AI Security Policy Competes With States but Cooperates With Private Firms

What the analysis examined

The piece is part of the AI Policy Corner series, a collaboration between the Montreal AI Ethics Institute (MAIEI) and the Governance and Responsible AI Lab (GRAIL) at Purdue University. It was written by Ogadinma Enwereazu, a PhD student in Political Science and a Graduate Affiliate at GRAIL.

The author drew on government documents published between 2019 and 2026, including authorization acts, executive orders and Department of Defense directives. The aim was to see how U.S. institutions frame AI in terms of strategic competition and cooperation.

The starting point was two developments from July 2025. The Department of Defense's Chief Digital and Artificial Intelligence Office (CDAO) announced contractual agreements with four frontier AI companies, Anthropic, Google, OpenAI and xAI, each worth up to $200 million. The same month, the White House's AI Action Plan declared it a "national security imperative" for the United States to achieve "unquestioned and unchallenged global technological dominance" in AI.

Competition against states, cooperation with industry

According to the author, the documents are more likely to discuss states when they discuss AI competition. Nearly three-quarters of competitive framings concern military advantage or rivalry with other states. Cooperation, by contrast, mostly involves partnerships with the private sector and industry.

The author says this is not entirely surprising. The Stanford HAI 2026 AI Index Report found that industry produced more than 90% of notable frontier AI models in 2025.

Government partnership with industry has precedents, such as NASA's extensive reliance on industrial contractors during the Apollo program and federally funded research that supported much of modern computing. What differs with AI, the author argues, is the direction of dependence. Earlier, the state took the early risks and firms commercialized the results. Today the most capable AI systems are first built by a relatively small number of companies, so the government depends on industry both to commercialize publicly supported innovations and to supply some of the most capable systems.

Three concerns

The author identifies three main concerns.

• Accountability. Treaties such as the North Atlantic Treaty require Senate approval, and executive agreements must be reported to Congress and in most cases published. Partnerships with companies may use contracts and flexible procurement deals whose details are often treated as confidential business information. The author notes that the GAO has found the Pentagon cannot tell whether many of its prototype agreements ever led to production or delivered capabilities to troops.
• Divided interests. Frontier AI companies answer to investors, global customers and their own commercial strategies. Their interests may often align with U.S. security goals, but not always. The author asks how the government can keep private commercial decisions consistent with national security priorities, and how much practical influence companies hold over security policy.
• Marginalized allies. If companies are treated as the primary AI partners, allied governments may be pushed to the margins on decisions that affect them. The author says this could complicate efforts to set shared international rules and standards for military AI.

What it means

The author concludes that as Washington increasingly competes with states while depending on private firms, the implications reach beyond government institutions. They extend to the companies developing these technologies and to the public, whose security and democratic accountability are at stake.

The piece is an analysis and opinion contribution rather than a policy announcement, but it points to open questions about oversight of defense AI contracts, such as those with Anthropic, Google, OpenAI and xAI, and about how allies fit into a company-centered approach.