US Security Agencies Must Work With Multiple AI Providers

Catenaa, Saturday, June 06, 2026 – US national security agencies must prioritize working with multiple AI providers, according to a memo from President Trump.

The memo was issued following an extended feud between the Pentagon and Anthropic, which until recently was the only vendor approved for classified military use.

The memo, signed by Trump on Friday, seeks assurances that government operations won’t be interrupted once an AI tool is embedded in national security systems, directing agencies to end work with companies that try to subvert the chain of command.

Yet it stops short of ordering that those firms be designated a supply-chain risk and does not require agencies to adopt the Pentagon’s standard of allowing all lawful uses of the technology, language that lay at the heart of the bitter showdown with Anthropic over the company’s insistence on extra safety measures.

Months of negotiations between the company and the Defense Department collapsed in late February, leading Pentagon officials to label Anthropic a supply-chain risk and move to abandon it as an AI provider for the armed forces. 

At issue in the dispute were Anthropic’s demands for restrictions barring the military from using the company’s tools in combat without human oversight and prohibiting AI-powered surveillance.

Trump issued the directive days after signing an executive order intended to address concerns that AI-enabled tools could be used for cyberattacks, part of a broader administration effort to shape policy on the emerging technology. 

On Friday, the President told reporters he was open to the idea of the US government owning stakes in major AI labs and planned to discuss it with executives from the companies as soon as next week.

In the works for months, the memorandum was intended to replace a previous one from the Biden administration on AI and national security. 

Some Trump administration officials have seen the new document as potentially offering political cover for the Pentagon to quietly back down on its claims that Anthropic posed a supply chain threat while diversifying its AI vendors, Bloomberg reported earlier.

Portions of the memo address some of the concerns raised during the Anthropic dispute, including a call for agencies to accelerate AI adoption while “safeguarding the constitutional chain of command.” 

That phrasing acknowledges the Defense Department’s concerns that an AI company could seek to have a veto over military workflows and Anthropic’s fears that the government could use AI models in ways that violate constitutional rights.

Agencies must not develop or use AI technologies to censor free speech, conduct “unlawful or unauthorized surveillance” activities or embed ideological bias, the memo makes clear, adding that AI adoption should be guided by “deep, proactive” partnerships with the AI industry.

Beyond surveillance, the memo also addresses Anthropic’s other main concern about preventing AI weapons systems that do not have human support by directing Defense Secretary Pete Hegseth to update a Pentagon directive on autonomous weapons systems to account for the “rapidly evolving capabilities of AI.”

The document calls for government entities, including the Central Intelligence Agency, Defense Department, and the Treasury Department, to leverage cutting-edge capabilities from diverse AI suppliers, while seeking assurances that those systems cannot be disabled, modified, or “severed in times of conflict.” 

It directs agencies to stop all business, even subcontracts, with companies that violate these terms, including firms that seek to pull the plug on use of their technology without government approval.

While the Pentagon is moving to wind down its use of some Anthropic tools for military uses, the company’s Mythos product, which has demonstrated a knack for finding vulnerabilities in computer systems, continues to be employed elsewhere in the government. 

The National Security Agency has been using Mythos to find weaknesses in popular software, including Microsoft products.

Anthropic is challenging the Pentagon’s supply-chain risk designation in court. It won a preliminary injunction in California but failed to secure a stay in a parallel case in Washington.