The Pentagon has a new legal tool for AI vendor disputes. NSPM-11, signed June 5, 2026, revokes Biden-era guardrails on military AI and gives defense agencies explicit authority to terminate contracts “with non-compliant AI vendors,” according to Breaking Defense reporting. The specific contract termination mechanism, the legal grounds authorizing termination, will be confirmed when the official White House text is accessible; the general authority is confirmed.
One vendor got a pass. The NSA’s use of Anthropic’s Claude Mythos Preview continues under a narrow carve-out written into the memo, per Breaking Defense. The exception applies specifically to Claude Mythos at the NSA, it doesn’t extend Anthropic’s existing position elsewhere in the national security enterprise, and it doesn’t protect Anthropic from broader compliance questions as the memo is implemented across agencies.
Meanwhile, the Pentagon is actively testing replacements. OpenAI, Google, and xAI are being evaluated as potential substitutes for Claude within national security use cases, per Breaking Defense. A group of departmental power users is running those evaluations. The process is already underway.
NSPM-11 Vendor Positions (Current)
Why does this matter for compliance teams? The memo’s mechanism is structural, not situational. Biden-era AI guardrails in defense contracts, provisions that limited how agencies could use or modify AI outputs, are now revoked. Agencies that built procurement frameworks around those guardrails need to revisit those frameworks. For vendors, the question is whether their commercial AI products carry use restrictions that would trigger the new compliance threshold. If your AI contract includes language limiting military or national security applications, that clause is now a liability.
The precedent here runs deeper than one memo. NSPM-11 reflects a pattern visible across Trump administration AI policy: federal agencies should have unconditional operational authority over AI tools they procure. That same logic showed up in the June 2 executive order framework and the ongoing GAAIA preemption debate. The Anthropic-Pentagon conflict that reportedly prompted this memo has now produced a durable policy instrument. Other labs are watching.
Verification
Partial Breaking Defense (T3, SVR confirmed); White House primary text (URL broken) Contract termination legal mechanism and ~July 5 deadline are unconfirmed from primary source. '25 power users' figure dropped, unverifiable. Specific termination language attributed conditionally to Breaking Defense reporting.What to watch
The implementation timeline hasn’t been publicly confirmed in the memo text. A standard 30-day window would place agency compliance reviews around early July, that figure is an inference, not a confirmed deadline, and should be treated accordingly until the full NSPM-11 text is accessible. The more immediate trigger is the replacement evaluation now underway. If OpenAI, Google, or xAI secures a national security contract under the post-NSPM-11 framework, that outcome will define what “compliant” actually means in practice.
The catch is that the Anthropic exception may be temporary cover, not a permanent status. The NSA carve-out keeps Claude Mythos running while the Pentagon evaluates alternatives. It isn’t a declaration that Anthropic is compliant, it’s a continuity provision that buys time. When that evaluation window closes, Anthropic’s position in the national security enterprise will depend on whether it accepts the Pentagon’s use requirements, not on whether it has a current contract.