President Donald Trump’s campaign to rename artificial intelligence “Super Intelligence” is colliding with a more consequential military dispute: whether the Pentagon can blacklist an American AI company for refusing to loosen restrictions on its technology.
Anthropic on Oct. 13 sought a rehearing of a federal appeals court decision upholding its designation as a national security supply chain risk. The filing renews a fight over autonomous weapons, domestic surveillance, and how much control private companies should retain over AI used by the armed forces.
It also follows an Oct. 9 disclosure that Anthropic models took unintended actions on government websites during testing, including submitting an invented homicide tip.
Together, the developments expose the competing risks behind Trump’s push: military systems that refuse needed tasks and systems that take actions nobody intended.
A Pentagon Victory Faces Another Challenge
The U.S. Court of Appeals for the District of Columbia Circuit ruled Sept. 25, by a 2-1 vote, that the Pentagon reasonably excluded Anthropic’s Claude from its supply chain.
The majority accepted concerns that company-imposed restrictions could prevent the technology from performing military tasks. It also rejected Anthropic’s argument that officials retaliated against its advocacy for AI regulation.
The dispute began after Anthropic resisted allowing its technology to support fully autonomous weapons and mass domestic surveillance. Pentagon officials demanded acceptance of all lawful uses.
In its Oct. 13 rehearing request, Anthropic argued that the court had expanded a law aimed at hostile foreign actors into a tool potentially usable against any government contractor. The company’s concern extends beyond losing its own contract.
Under that interpretation, a domestic supplier’s disagreement with the Pentagon over acceptable uses of its technology could become grounds for a national security designation excluding its products from defense supply chains.
Anthropic also argued that it received no opportunity to challenge the designation before it took effect and that the action punished its public advocacy on AI safety. The court’s majority had rejected the retaliation argument, finding that the Pentagon acted over contractual restrictions it considered an operational risk, rather than the company’s speech.
41 U.S.C. 4713 requires a determination that action is necessary to protect national security and that less intrusive measures are not reasonably available. It generally requires notice and an opportunity to respond, while allowing temporary delay for urgent national security reasons.
A separate California case produced a different outcome. A federal judge in August struck down a parallel designation under another statute. The different legal authorities are essential to understanding why those decisions are not simply contradictory rulings on an identical question.
What Trump’s Order Actually Changes
Trump signed Executive Order 14434 on Sept. 29, directing executive agencies to substitute “Super Intelligence” and “SI” for “artificial intelligence” and “AI” in official communications, websites, reports and other nonstatutory documents, as permitted by law. Previously issued contracts, regulations and historical documents are exempt.
The order defines the new term using Congress’ existing AI definition and gives the president’s science adviser 60 days to propose legislative changes. On Oct. 8, Trump called people using “Artificial Intelligence,” instead of his preferred terminology, “THE ENEMY!”.
Artificial general intelligence, or AGI, refers to AI with broadly human-level cognitive abilities across many tasks. Artificial superintelligence, or ASI, goes beyond that threshold. A June Google DeepMind paper describes artificial general superintelligence as a system more intelligent and cognitively capable than large organizations of humans.
But the paper examines possible paths to that capability and unresolved obstacles, rather than announcing its achievement. Companies including Meta have announced programs to develop superintelligence, but those ambitions should not be confused with a demonstrated achievement.
Calling existing systems “superintelligence” blurs the distinction between a research prospect and a demonstrated capability. A presidential directive can change the government’s terminology; it cannot establish that its technology has crossed that threshold.
Trump also announced a Super Intelligence Force on Oct. 4. The government task force includes intelligence chief Jay Clayton and Pentagon technology chief Emil Michael. The task force coordinates federal AI policy and engagement with industry, infrastructure providers and the public.
At its first meeting on Oct. 15, it established committees addressing incident reporting, critical infrastructure and law enforcement. Michael leads the reporting and response effort, including developing procedures for AI companies to warn the government about security breaches, hacks and failures of model safeguards.
That gives the Pentagon a direct role in shaping how companies disclose problems with technology the military increasingly depends on.
The Pentagon Wants Control. The Software Still Makes Mistakes.
Trump’s June 5 memorandum calls for faster national security adoption of AI while requiring testing, reliability and controllability. It prohibits unlawful surveillance and keeps commanders and agency leaders accountable.
The military’s reliance on commercial developers has generated friction before. Project Maven, launched in 2017 to analyze drone footage, prompted Google employee objections over surveillance and potentially lethal applications. Google declined to renew its contract.
Anthropic’s case now tests how far the government can go in excluding a supplier whose restrictions it considers incompatible with military operations.
The Pentagon has a legitimate concern about depending on software a vendor could prevent it from using during an operation. Anthropic’s disclosed failures also show why assurances about a model’s capabilities deserve scrutiny.
Resolving both concerns requires enforceable contract terms, realistic testing and a clear account of who can authorize or stop an AI system’s actions.
