The Pentagon said it would give warfighters the tools they need to act with confidence.
In a moment that may mark a turning point in the relationship between Silicon Valley and the state, the Pentagon has formalized agreements with eight major technology firms to weave artificial intelligence into the fabric of military operations. The arrangement is framed as a safeguard against over-reliance on any single vendor, yet it also reveals a deeper tension: one prominent AI company, Anthropic, was left out precisely because its leadership dared to name the risks aloud. What unfolds now is less a story about contracts than about who gets to define the boundaries of machine-assisted power — and who bears the cost of asking that question.
- The Pentagon is moving faster than public debate can follow, embedding AI across defense operations for over a million military users while the ethical frameworks to govern those tools remain unresolved.
- Anthropic's exclusion from the contracts — a direct consequence of its CEO publicly warning against autonomous weapons and mass domestic surveillance — signals that candor about AI risk carries a real institutional price.
- OpenAI stepped into the gap almost immediately, and Google's Gemini will now handle classified government work for the first time, even as hundreds of Google's own researchers signed a letter urging the company to pull back.
- The Pentagon's deliberate spread of contracts across eight firms — Google, OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and Reflection — is designed to prevent vendor dependency, but it also distributes accountability so widely it may dilute it.
- Anthropic has filed a legal challenge to its exclusion, with a court date set for September that could either redraw the boundaries of defense-AI partnerships or simply ratify the direction already chosen.
The Pentagon has signed contracts with eight technology companies — Google, OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and the startup Reflection — to embed artificial intelligence across military operations. Officials described the multi-vendor strategy as a deliberate effort to avoid over-dependence on any single provider, while giving defense personnel access to a broad range of AI capabilities. The military's own AI platform, launched last year, has already reached more than a million users, with personnel reporting that tasks once measured in months can now be completed in days.
The roster's most conspicuous feature is who is missing. Anthropic, the AI safety company, was excluded after its chief executive Dario Amodei publicly raised concerns about the potential use of advanced AI for mass surveillance of American citizens and for autonomous weapons systems operating without human oversight. Though Anthropic's tools — including versions of its Claude chatbot — remain in use across some government contexts, and the company was the first AI firm cleared to handle classified information, the relationship with the Pentagon has fractured. Anthropic has filed a legal challenge to its exclusion, with the case expected in court this September.
OpenAI moved swiftly to fill the space, becoming the first company to sign a new Pentagon deal following Anthropic's public break. Google's Gemini, already present in some government work, will now be used on classified material for the first time — a step that prompted hundreds of Google employees, including researchers from DeepMind, to write to CEO Sundar Pichai urging restraint. Google declined to comment. SpaceX will deploy Grok, the chatbot from Elon Musk's xAI, though it is widely considered less capable than its rivals. Nvidia and Reflection will contribute open-source models, while Microsoft, Amazon, and Oracle extend their existing cloud infrastructure roles to support AI deployment at scale.
The Pentagon's language — authorizing AI for 'lawful operational use' across any military purpose — leaves the practical boundaries of these tools deliberately wide. With Anthropic's legal challenge pending and the deeper questions about autonomous weapons and surveillance still unanswered, September's court date may either reshape the landscape of military AI or simply confirm the course already set.
The Pentagon has signed eight new contracts with major technology companies to embed artificial intelligence across military operations, marking an aggressive expansion of what officials are calling an 'AI-first fighting force.' The deals involve Google, OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia, and the startup Reflection—a roster that notably excludes Anthropic, the AI safety company whose leadership has become increasingly vocal about the risks of deploying advanced AI tools in warfare and domestic surveillance.
The Pentagon framed the multi-vendor approach as a deliberate strategy to avoid what it calls 'vendor lock,' the vulnerability of depending too heavily on a single company for critical technology. By spreading contracts across eight firms, the department said it would give military personnel access to a diverse set of AI capabilities while maintaining resilience and competition. The announcement came as the military's own AI platform, launched last year, has already reached more than a million users across the defense department—personnel who have reported cutting the time required for routine tasks from months down to days.
Anthropicís absence from the new contracts reflects a deepening rift between the company and the Pentagon. Earlier this year, Anthropic's chief executive Dario Amodei went public with concerns that defense agencies could use powerful AI systems to conduct mass surveillance of American citizens and to deploy fully autonomous weapons systems without human oversight. The company's tools, including versions of its Claude chatbot, remain in use across various government and defense agencies—Anthropic was the first AI company cleared to work with classified information—but the relationship has fractured. The company has filed a legal challenge to the Pentagon's decision to exclude it, with the case expected to reach court in September.
OpenAI moved quickly to fill the opening, becoming the first company to sign a new Pentagon deal after Anthropic's public break. A company spokesperson said the firm believed American military personnel deserved access to the world's best tools. Google's Gemini, already deployed in some government contexts, will now be used to handle classified work for the first time. The expansion prompted hundreds of Google employees, including researchers from the company's DeepMind division, to send a letter to chief executive Sundar Pichai urging the company not to deepen its military partnerships. Google declined to comment on the letter or the new contracts.
The remaining signatories bring different capabilities and histories to the arrangement. SpaceX, now the parent company of Elon Musk's AI startup xAI, will deploy the chatbot Grok, though the system is widely regarded as less sophisticated than competitors like Anthropic or OpenAI. Nvidia and Reflection will contribute open-source AI models—Nemotron and Reflection 70B respectively—though Nvidia is not providing hardware as part of the deal. Microsoft, Amazon Web Services, and Oracle have long provided the cloud infrastructure that enables classified government work online; the new contracts extend that role to support the deployment of AI models at scale.
The Pentagon's move reflects a broader acceleration in military AI adoption, one that has outpaced public debate about the technology's implications. The department's language—'lawful operational use' for any military purpose—leaves considerable room for interpretation about what kinds of decisions and actions AI systems might inform or execute. With Anthropic's legal challenge pending and the company's concerns about autonomous weapons and mass surveillance still unresolved, the question of how these tools will actually be used in practice remains open. The September court date could reshape the landscape of defense-AI partnerships, or it could simply formalize the Pentagon's current direction.
Citações Notáveis
These agreements accelerate the transformation of the US military as an AI-first fighting force— Pentagon statement
We believe the people defending the United States should have the best tools in the world— OpenAI spokesperson
A Conversa do Hearth Outra perspectiva sobre a história
Why did Anthropic step back when the others stepped in? It seems like a lucrative contract.
Anthropic's leadership decided the risk to their values was higher than the revenue. Dario Amodei saw a path toward autonomous weapons and domestic surveillance, and he chose to say so publicly rather than quietly accept the work.
But Anthropic's tools are still being used by the Pentagon right now.
Yes—they were the first company cleared for classified work, so they're embedded in the system. But the new contracts are the future. Anthropic wanted a say in how that future looked. The Pentagon said no.
What does 'lawful operational use' actually mean in practice?
That's the question nobody wants to answer yet. It's broad enough to cover almost anything the military decides is lawful. That's partly why Anthropic fought back.
Will the court case in September change anything?
It could. If Anthropic wins, it might force the Pentagon to reconsider how it deploys these tools. If it loses, the current direction becomes the settled law of military AI.
Why did Google's employees write that letter?
They understood something their executives were willing to ignore: once you build the tool, you don't control how it gets used. The letter was a last attempt to slow that down.