US federal conflict with Anthropic and tightening of AI procurement and access rules

In 2026, the US federal government's approach to artificial intelligence (AI) deployment is undergoing a significant transformation, marked by increased scrutiny, tighter regulation, and strategic efforts to secure domestic AI infrastructure. Central to these developments is the escalating conflict with Anthropic, a major AI provider, which has been officially designated as a supply chain risk by the Pentagon—an action that underscores the administration’s focus on national security and control over critical AI assets.

Pentagon’s Designation of Anthropic as a Supply-Chain Risk

The Pentagon has formally notified Anthropic that it considers the company—and its flagship AI models—a "supply chain risk". This designation follows a series of disputes over AI safety, security, and compliance with federal standards. Notably, Anthropic has been involved in ongoing discussions with the Department of Defense (DoD) after the company’s AI models, such as Claude, were reportedly used in sensitive regions like Iran, raising concerns about oversight and secure deployment.

Anthropic’s CEO, Dario Amodei, has publicly challenged the Pentagon’s risk assessment, even filing a lawsuit against the department to contest this designation. The Pentagon’s move reflects a broader trend of tightening control over AI tools used by federal agencies, emphasizing security, interoperability, and sovereignty in AI procurement and deployment.

Furthermore, there have been threats by the Pentagon to invoke the Defense Production Act (DPA) against Anthropic, signaling a willingness to leverage emergency powers to ensure the security and resilience of AI supply chains. This stance illustrates the heightened geopolitical and security concerns surrounding AI, especially when foreign or domestically scrutinized companies are involved.

New Strict Guidelines and Contract Terms Reshaping Federal AI Procurement

In response to these tensions and the evolving security landscape, the US federal government is implementing strict new guidelines for AI procurement. These guidelines aim to standardize contract terms to ensure "any lawful use" of AI models in government operations, while emphasizing security, transparency, and control.

Key features of these new policies include:

• Enhanced vetting and security protocols for AI vendors, prioritizing domestic and trusted providers.
• Mandatory compliance checks aligned with sector-specific guardrails—particularly in sensitive sectors such as defense, healthcare, and urban infrastructure.
• Open-source and transparent platform adoption, exemplified by moves like the State Department moving its StateChat to GPT-4.1, promoting flexibility while maintaining security standards.
• The adoption of trust infrastructure measures such as digital signatures and content provenance systems to verify authenticity and prevent misinformation.

The General Services Administration (GSA) and other federal agencies are also revising their AI strategies to reflect these tighter rules. This includes shifting towards mixed procurement models that favor open-source models like NVIDIA’s NemoClaw and OpenAI’s GPT-4.1, which offer a balance of security, transparency, and scalability.

Broader Context: Sovereign Infrastructure and Regulatory Divergence

These procurement reforms are part of a larger strategy to develop domestic AI infrastructure and reduce reliance on foreign providers. Countries like the UK, Japan, and South Korea are heavily investing in sovereign AI compute centers and secure AI chip manufacturing to bolster data sovereignty and national security.

However, at the US federal level, the proliferation of subnational regulations—such as Texas’ Responsible AI Act or California’s incident reporting mandates—has created a fragmented regulatory landscape. This divergence complicates interoperability, especially for agencies operating across multiple jurisdictions, and underscores the importance of federal standards to ensure security and consistency in AI deployment.

Enhancing Security, Governance, and Trust

As AI systems become integral to critical government functions, the emphasis on security and governance intensifies. Techniques such as federated learning and secure multi-party computation (SMPC) are increasingly adopted to facilitate privacy-preserving data sharing. Simultaneously, organizations are deploying deepfake detection tools, content verification systems, and digital signatures to maintain public trust and content integrity.

The security of AI supply chains is also a priority; the designation of Anthropic as a risk highlights concerns over model poisoning, prompt injection, and data leakage. The government’s focus on red-team exercises and model integrity checks aims to preempt malicious exploitation.

Conclusion

The US federal government’s evolving stance in 2026 reflects a deliberate effort to tighten control over AI supply chains, establish stringent procurement rules, and foster secure, domestically-controlled AI ecosystems. The conflict with Anthropic exemplifies the broader drive to protect national security interests and ensure trustworthy AI deployment in critical sectors.

Moving forward, the challenge lies in balancing innovation with security and interoperability. Achieving this will require international cooperation, harmonized standards, and resilient security frameworks to ensure that AI remains a trusted enabler of public service, rather than a source of fragmentation or vulnerability. The strategic emphasis on secure infrastructure, trust measures, and workforce upskilling will be pivotal in shaping the responsible evolution of public-sector AI in the years to come.