Skip to content
Matterfile

American Frontier AI Firms Should Cease All Cooperation With the US Military

Government-side notes arguing that American frontier AI firms should stop cooperating with the US military. Four arguments: frontier AI lowers the cost of intervention and removes human hesitation, it speeds up a US–China arms race and regulatory capture, it pulls research away from socially productive uses, and it weakens democratic checks.

Author
Chris Hu
Side:
Government
Length
5 min read

Motion

THBT American Frontier AI Firms Should Cease All Cooperation with the US Military

Case notes. These are working notes prepared for a debate round rather than a finished essay.

Editor's note: Spelling and typing errors have been corrected and headings formatted for the web.

Background

Background as it appears in the prep document.

The Anthropic–Pentagon dispute, 2026. Anthropic resisted Pentagon demands over AI use, particularly concerning autonomous weapons and mass domestic surveillance. In February 2026, the Trump administration responded by seeking to exclude Anthropic from government and defence business. Subsequent court rulings limited some of the government's actions, while allowing other restrictions to remain possible.

Palantir's Maven Smart System. The US military uses AI to process intelligence from surveillance systems and identify potential military targets. In March 2026, the Pentagon moved to institutionalize Palantir's Maven as a core military system.

The Minab school strike in Iran, February 2026. A September 2026 Bloomberg investigation reported that flawed intelligence, outdated imagery and overreliance on AI contributed to a strike that killed civilians, including children.

Military AI already exists outside frontier labs. Companies such as Palantir, Anduril, Shield AI and Scale AI develop defence-specific AI software, platforms and autonomous systems. Some rely on foundation models developed elsewhere.

Argument 1: Frontier AI increases violence and casualties

Giving the US military frontier AI increases its capacity to conduct harmful operations in two ways:

  1. The military uses AI to more effectively pursue undesirable objectives
    1. The cost of intervention reduces: Traditionally, intelligence collection, target identification and operational planning require extensive human resources. Governments have to invest enormous amounts of time and personnel to pursue a campaign. AI is able to automate this portion of the work. Even if individual strikes become more precise, a massive increase in the number of strikes causes more deaths.
    2. AI removes human resistance in decision making: Many military generals oppose military operations because A) humans experience empathy, if you watch footage of civilians near a targeted site you feel personally guilty for bombing them. B) People care about accountability for making the wrong decision. Given that AI lacks moral discomfort, they are more inclined to cause innocent casualties.
    3. Surveillance becomes worse: AI is incredibly capable at synthesizing different types of data, i.e. AI can centralize information from different sources, such as text, image and audio, very efficiently. AI can detect patterns across large amounts of datasets. It also makes surveillance more scalable so you monitor more people. This makes governments more efficient at cracking down protests and arresting activists preemptively.
  2. Frontier models can make military mistakes particularly dangerous:
    1. Capability Problem:
      1. General-purpose models are typically trained to perform general tasks; they are not designed for military solutions. So frontier models are oftentimes just quite bad at checking if the imagery is current or the civilians are nearby. This is why the Minab school strike was directly caused by incompetent AI using outdated imagery.

        Editor's note: the background section above describes the Bloomberg investigation as reporting that flawed intelligence, outdated imagery and overreliance on AI contributed to the strike.

      2. The obvious opp response is that the military would simply try to develop their own AI, but:

        1. We win on a time delay. Front tier models are computational better and more capable, so it takes more time on Opp for the government to invest and fund alternative AI model developments.

Argument 2: Existential risk and AI arms race

  1. Militarization intensifies the US–China AI race
    1. China feels less secure when they see AI integration in the US military because it’s unclear if the US will use AI offensively or defensively. This security dilemma incentives accelerated AI development by China. This makes alignment testing and model release restrictions less desirable because this might give the other country a massive first mover advantage.
  2. Military competition means countries gatekeep information
    1. You are less likely to enter into common safety standards with other countries because doing so might reveal your model’s limitations or capability.
    2. Regulatory Capture: The government becomes dependent on precisely the industry that it is supposed to regulate. You have no incentives to regulate the AI industry when your military is heavily dependent on its outputs. The military itself is gonna lobby against AI regulations.
  3. Opp will talk about Trump’s regulatory backlash if AI firms stop cooperation, this is good:
    1. Government backlash and reduced investor confidence will reduce capital invested to train a more advanced system. This gives society more time to develop regulatory institutions and safety evaluations.

Impacts

Faster development of AI creates many existential risks:

  1. The military will allocate more long sequence tasks to AI, giving them greater control over software and digital infrastructure. This makes the alignment problem more likely to occur because it’s easier for them to strive for misaligned proxy goals in more complicated tasks.
  2. More capable AIs will also be more likely to be used by malicious actors to develop biological weapons or other misuse.

Argument 3: Military cooperation distorts AI research away from socially productive uses

  1. Military contracts change what companies optimize for:
    1. Companies naturally respond to their most commercially valuable customers because they are profit oriented. If the military becomes an important customer, AI companies will increasingly invest in capabilities relevant to military procurement, i.e. intelligence analysis and specialized cyber capabilities. These investments compete with alternative research directions such as education and scientific research.
    2. Military technology is subject to confidentiality and security requirements. This means:
      1. First, independent researchers have fewer opportunities to scrutinize claims about safety because they have less access to all the information.
      2. Second, talented international researchers will be excluded from projects because of nationality or potential spy activities.
      3. Third, research becomes more fragmented as groups duplicate work that cannot be openly shared.
    3. Military funding changes who enters the AI industry: Because military AI is being funded, this creates path dependency. Universities build relationships with defence contractors. Research labs specialize in defence applications.
  2. Other socially productive use of AI are more valuable:
    1. AI used in education or administrative work can benefit millions of people. Military AI kills people and generates arms races.

Argument 4: Democracy

  1. Tech giants in the US become far less likely to criticize the government:
    1. You are dependent on government contracts, so you avoid criticizing the government. Technology executives like Elon Musk possess substantial influence over public debate and have large social followings.
  2. Tech companies will lobby for militarization:
    1. Once military procurement becomes a revenue opportunity, companies lobby for higher defense budgets and faster procurement.
  3. Frontier AI lowers the domestic political cost of war:
    1. One reason governments hesitate to go to war is the political cost of exposing their own soldiers to danger. Military AI reduces the involvement of physical soldiers and personnel for intelligence roles.

Sources

  1. Bloomberg investigation into the Minab school strike (September 2026)Referenced in the document's background section; no title, author or link given.
  1. ArgumentTechnology & AI, Economics, Politics21 min read

    AI Does, or Will Do, More Harm Than Good

    A Proposition case arguing that AI does or will do more harm than good, framed as a normative debate rather than a policy debate about banning AI. It makes five arguments: existential risk, labour-market displacement and inequality, threats to democracy and escalation of conflict, structural bias against marginalized groups, and mass invasions of privacy.

  2. ArgumentPolitics, Global Development12 min read

    Attitudes Before Laws: How Progressive Movements in Developing Countries Should Campaign

    An Opposition case on whether progressive movements in developing countries should campaign for legal change or attitudinal change. It argues that changing attitudes is both better (laws follow attitudes and go unenforced without social buy-in) and easier (governments are often corrupt, captured and coalition-bound, while individuals can be persuaded).