Pentagon strikes classified AI deals with OpenAI, Google, and Nvidia — but not Anthropic

The News

The United States Department of Defense (DoD) has entered into classified agreements with OpenAI, Google, and Nvidia, securing access to advanced AI capabilities while excluding Anthropic [1]. The deals, announced in early May 2026, mark a significant shift in the Pentagon’s AI procurement strategy, signaling a move toward closer partnerships with key technology providers [2]. While contract specifics remain classified, reports indicate the agreements involve access to advanced large language models (LLMs), specialized AI infrastructure, and potential custom model development tailored for defense applications [1]. This development follows heightened scrutiny of the DoD’s reliance on AI vendors and a recent dispute with Anthropic over usage terms [2]. The agreements with Google and OpenAI are particularly noteworthy, given internal dissent within Google regarding its involvement in classified military AI projects [3].

The Context

The DoD’s decision to partner with OpenAI, Google, and Nvidia highlights a complex interplay of technological necessity, strategic risk mitigation, and evolving ethical considerations. OpenAI, developer of the GPT family of LLMs, DALL-E text-to-image models, and Sora text-to-video models, holds a unique position in the AI landscape [1]. These models have demonstrably influenced industry research and commercial applications, making them attractive for defense applications requiring advanced natural language processing and generative capabilities [1]. Google, with its DeepMind division, represents another critical partner, bringing substantial expertise in AI research and development, particularly in reinforcement learning and complex problem-solving [4]. Nvidia’s involvement underscores the role of specialized hardware in AI deployment; the company’s GPUs are the de facto standard for training and inference of large AI models [1]. The deals with Microsoft and AWS, as reported by TechCrunch [2], further emphasize the DoD’s strategy of diversifying its AI infrastructure across multiple cloud providers, reducing reliance on any single vendor.

The exclusion of Anthropic is arguably the most significant element of this announcement. The decision is likely linked to a recent, undisclosed dispute between the DoD and Anthropic regarding the terms of service for Anthropic’s Claude LLMs [2]. While the specifics remain confidential, the disagreement likely involved concerns about data usage, intellectual property rights, or restrictions on the DoD’s ability to modify or adapt the models for military purposes. This dispute has prompted the DoD to actively diversify its AI vendor portfolio, mitigating dependency on a single provider [2]. The timing of this diversification is significant, occurring alongside Elon Musk’s public accusations against OpenAI CEO Sam Altman and president Greg Brockman, alleging deception regarding the company’s original mission and revealing that Musk’s new company, xAI, is effectively distilling OpenAI’s models [4]. Musk’s trial against OpenAI, which included revelations about his $38 million initial investment and the company’s eventual valuation reaching $800 billion, then $1 trillion, and ultimately $1.75 trillion, has further complicated the landscape of AI development and government partnerships [4].

The technical architecture underpinning these partnerships is likely to involve a hybrid approach. The DoD will leverage OpenAI and Google’s cloud-based AI services, potentially establishing secure enclaves within those environments for classified data processing [1]. Nvidia’s GPUs will be deployed both within these cloud environments and on-premise at DoD facilities to accelerate AI workloads [1]. The use of specialized AI frameworks like NVIDIA’s NeMo, a scalable generative AI framework written in Python and boasting 16,855 stars on GitHub, is highly probable, enabling the DoD to customize and optimize AI models for specific military applications. Deployment will require robust cybersecurity measures and stringent data governance protocols to protect sensitive information and prevent unauthorized access [1].

Why It Matters

The DoD’s classified AI deals have far-reaching implications for developers, enterprise AI adoption, and the broader AI ecosystem. For AI engineers, these agreements create both opportunities and challenges. The prospect of working on advanced AI projects with significant real-world impact is appealing, but it raises ethical concerns about potential misuse of AI technology [3]. The demand for specialized expertise in secure AI development, adversarial machine learning, and explainable AI is likely to rise [1].

For enterprises and startups, the DoD’s increased investment in AI represents a significant market opportunity but also intensifies competition. Companies seeking DoD contracts will need to demonstrate technical excellence and a commitment to ethical AI practices and robust data security protocols [2]. The exclusion of Anthropic underscores the importance of aligning business models with government priorities and proactively addressing potential conflicts of interest [2]. Compliance with DoD regulations and security requirements can be costly, potentially creating a barrier to entry for smaller companies [2]. The ongoing debate at Google, evidenced by a letter signed by over 600 employees demanding the Pentagon block its AI models [3], highlights the growing tension between commercial interests and ethical considerations in the AI industry.

The winners in this ecosystem are likely to be companies providing secure, reliable, and ethically aligned AI solutions. OpenAI, Google, and Nvidia are well-positioned to benefit from these deals but also face increased scrutiny and pressure to ensure responsible AI development [1]. The losers are those failing to meet DoD requirements or perceived as posing unacceptable ethical risks [2]. The controversy surrounding Anthropic serves as a cautionary tale for other AI vendors seeking government engagement [2].

The Bigger Picture

The DoD’s classified AI deals represent a broader trend of global governments investing heavily in AI to enhance national security and maintain a competitive edge. This trend is accelerating as AI technology becomes more sophisticated and accessible [1]. China, for example, has made significant investments in AI research and development, actively exploring military applications of AI [1]. The US government’s focus on AI is also driving innovation in the private sector, as companies compete to develop next-generation AI technologies [1].

The DoD’s prioritization of partnerships with OpenAI, Google, and Nvidia over Anthropic signals a shift toward integrated AI procurement, moving away from transactional relationships toward long-term strategic collaborations [1]. This approach is likely to become more common as governments seek to leverage AI’s full potential to address complex national security challenges [1]. The ethical implications of military AI are likely to intensify as the technology becomes more powerful and autonomous [3]. The recent publication of a study by Grossman et al. analyzing the disruptive impact of generative AI on Google Search, Gemini, and AI Overviews further underscores AI’s transformative potential and the need for careful societal consideration [4].

The reliance on a small number of dominant AI providers creates a concentration of power and influence, raising concerns about vendor lock-in and anti-competitive behavior [1]. The DoD will need to actively monitor the AI landscape and diversify its vendor portfolio to mitigate these risks [2]. The emergence of open-source AI models, such as GPT-OSS-20B (6,929,145 downloads) and GPT-OSS-120B (4,083,858 downloads) on HuggingFace, offers alternatives to proprietary solutions, though these models often lack the performance and reliability required for critical military applications [1].

Daily Neural Digest Analysis

Mainstream media coverage of the DoD’s classified AI deals has focused on geopolitical implications and technological advancement. However, a critical element often overlooked is the inherent fragility of these partnerships. The internal dissent within Google regarding military AI use [3], coupled with Musk’s legal battle against OpenAI [4], creates a volatile environment that could disrupt these agreements at any time. The DoD’s reliance on a small number of vendors exposes it to operational and security risks. A major data breach at OpenAI or Google, for example, could compromise sensitive military information [1].

The exclusion of Anthropic, while strategically prudent, highlights a deeper tension between AI innovation and ethical constraints. The DoD’s prioritization of security and control over potentially more advanced models may stifle innovation and limit its ability to respond to evolving threats [2]. The rapid pace of AI development means current agreements may quickly become obsolete, requiring constant renegotiation and adaptation [1].

The question remains: can the DoD effectively manage the risks and complexities of these classified AI partnerships while fostering responsible AI development? The answer depends on the DoD’s ability to establish clear ethical guidelines, promote transparency, and cultivate a diverse AI vendor ecosystem. The current trajectory suggests a cautious but necessary embrace of AI, though the long-term success of this strategy remains uncertain.

---

References

[1] Editorial_board — Original article — https://www.theverge.com/ai-artificial-intelligence/922113/pentagon-ai-classified-openai-google-nvidia

[2] TechCrunch — Pentagon inks deals with Nvidia, Microsoft, and AWS to deploy AI on classified networks — https://techcrunch.com/2026/05/01/pentagon-inks-deals-with-nvidia-microsoft-and-aws-to-deploy-ai-on-classified-networks/

[3] The Verge — Google employees ask Sundar Pichai to say no to classified military AI use — https://www.theverge.com/ai-artificial-intelligence/919326/google-ai-pentagon-classified-letter

[4] MIT Tech Review — Musk v. Altman week 1: Elon Musk says he was duped, warns AI could kill us all, and admits that xAI distills OpenAI’s models — https://www.technologyreview.com/2026/05/01/1136800/musk-v-altman-week-1-musk-says-he-was-duped-warns-ai-could-kill-us-all-and-admits-that-xai-distills-openais-models/