Written by Kelly M. Saylor
On February 27, 2026, President Donald J. Trump directed federal agencies to “immediately cease all use.” [American AI company] Anthropic technology. ” Secretary of Defense Pete Hegseth (currently using “Secretary of the Army” as a “secondary title” pursuant to Executive Order (EO) 14347, dated September 5, 2025) subsequently directed the Department of Defense (DOD, currently using “Department of the Army” as a secondary title pursuant to EO 14347) to designate Anthropic as a supply chain risk to national security. Defense contractors, suppliers, and partners are prohibited from working with Anthropic. and a transition period of up to six months away from Anthropic products.
The designation reportedly follows a months-long dispute between the Department of Defense and Anthropic over DOD’s use of Anthropic products, including Claude, the company’s generative AI model. On March 9, Anthropic filed a civil suit challenging these directives in the U.S. District Court for the Northern District of California and a petition for review in the D.C. Circuit Court of Appeals. A district court issued a preliminary injunction in Anthropic’s favor on March 26, but the Court of Appeals denied Anthropic’s motion for an injunction on April 8, reversing the lower court’s injunction.
Some lawmakers are calling for Congress to act to resolve the differences and set rules for the department’s use of AI and autonomous weapons systems.
background
In July 2025, the Department of Defense announced that it had awarded contracts worth up to $200 million each to Anthropic, Google, OpenAI, and xAI to “accelerate the Department of Defense’s (DoD) deployment of advanced AI capabilities to address critical national security challenges.” Although the Department of Defense has not publicly disclosed the full range of use cases for these companies’ AI models, Anthropic notes that Claude is “reportedly the most widely deployed and used frontier AI model in the Department of Defense.” Anthropic further states that its models are used “across the Department of the Army and other national security agencies for mission-critical applications such as intelligence analysis, modeling and simulation, operational planning, and cyber operations.” Although Anthropic’s usage policy prohibits using models to incite violence or develop or design weapons, reports indicate that Claude was used in the January 2026 operation to capture Venezuelan President Nicolas Maduro.
According to reports, Anthropic inquired about the Department of Defense’s use of Claude, which resulted in concerns within the agency that Anthropic would disapprove of certain use cases and therefore seek to limit the DoD’s use of the model. As a result, the Department of Defense has reportedly asked Anthropic and other AI companies to allow the use of their AI models for “all lawful purposes.” Anthropic was reportedly “open to adapting its usage policy for the Department of Defense,” but the company was reluctant to allow two use cases: domestic mass surveillance and fully autonomous weapons systems, evaluating “what can be done safely and reliably with today’s technology.” Anthropic CEO Dario Amodei, in explaining his decision to deny the Department of Defense’s request for “full and unrestricted access to Anthropic’s models,” said autonomous weapons systems “could prove critical to our national defense. But today, Frontier AI systems are not reliable enough to power fully autonomous weapons.”
It is not publicly known that the DOD uses Claude (or any other Frontier AI model) within autonomous weapons systems. Department of Defense Directive (DODD) 3000.09, Autonomy of Weapon Systems, outlines the approval process for the development and fielding of autonomous weapons systems and identifies requirements for their use.
What is an autonomous weapon system?
DODD 3000.09 defines autonomous weapon systems as “weapon systems.”[s] Once activated, targets can be selected and attacked without further intervention. [a human] operator. “This concept of autonomy humans are outside the mosquito net or complete autonomy. The Directive contrasts such systems with human-supervised systems. human on loopautonomous weapon systems allow operators to monitor and stop weapons from attacking targets. Another category is semi-autonomous, or person in the loop“a weapon system that attacks only individual targets or specific groups of targets selected by” [a human] operator. ”
DoD 3000.09 requires that all systems, including autonomous weapons systems, be designed “to enable commanders and operators to exercise an appropriate level of human judgment in the use of force.” Such decisions do not require manual human “control” of the weapon system, but require broader human involvement in decisions about how, when, where, and why the weapon is used (i.e., humans need to assess the operational environment and decide to deploy a weapon that can operate autonomously). This commitment involves a human decision that the weapon will be used “with due care and in accordance with the laws of war, applicable treaties, weapons system safety rules, and applicable rules of engagement.” The requirement for “human judgment regarding the use of force” does not imply that such systems are operated with human involvement.
Related legislation and parliamentary issues
The Department updated DODD 3000.09 in January 2023, and later that year, Congress passed the National Defense Authorization Act for Fiscal Year 2024 (NDAA; PL 118-31). Section 251 requires the Secretary of Defense to notify the defense committees of changes to DODD 3000.09 within 30 days. The Secretary is directed to provide an explanation of the amendments and an explanation of the reasons for the amendments. Additionally, section 1066 of the NDAA for Fiscal Year 2025 (PL 118-159) requires the Secretary to annually submit to the Committee, through December 31, 2029, a “comprehensive report on the authorization and fielding of lethal autonomous weapons systems by the United States.” Congress has not legislated the department’s use of AI models or their reliability. Some Member States have introduced related legislation (e.g. S. 1394 and HR 2894, 118)th Conference; S. 4113, 119th meeting).
If Congress determines that additional oversight is needed, it may codify the requirements in DOD 3000.09 or consider additional notification requirements for DOD’s use of autonomous weapons systems and AI models. Congress could also limit funding for the development and/or use of autonomous weapons systems or certain use cases of AI models by the Department of Defense. Congress may restrict such uses if Congress determines that such uses pose an unacceptable level of risk at this stage of technology development.
- About the author: Kelly M. Saylor, Advanced Technology and Global Security Specialist
- Source: This article was published by the Congressional Research Service (CRS).
