Artificial intelligence is becoming more useful to the Pentagon, even as the companies building the technology insist their systems should not be used to harm people. The result is a fast-moving and uncomfortable middle ground: AI is not being described as a weapon, but it is already helping military leaders move faster through parts of the kill chain.
Dr. Radha Plumb, the Pentagon's chief digital and AI officer, told TechCrunch that AI is giving the Department of Defense a “significant advantage” in identifying, tracking and assessing threats. She also said the technology is helping commanders respond in time to protect U.S. forces.
What the Pentagon says AI is doing
The kill chain is the military process for finding, following and eliminating threats. It depends on a network of sensors, platforms and weapons, and it includes decisions made before any force is used.
According to Plumb, generative AI is proving useful in the planning and strategy stages of that process. That distinction matters. The Pentagon is presenting these tools as support for commanders rather than as systems that independently choose targets or fire weapons.
Plumb said: “We obviously are increasing the ways in which we can speed up the execution of kill chain so that our commanders can respond in the right time to protect our forces.”
She also described scenario planning as an area where generative AI can help. In her account, AI can assist commanders as they evaluate response options, use available tools and think through trade-offs when facing a threat or a series of threats.
That use case shows why the technology is attractive to defense agencies. Generative AI can organize information, compare possible paths and support faster planning. In a military context, however, planning can sit very close to decisions about force.
AI companies are entering defense more openly
The relationship between major AI developers and the Pentagon is still relatively new. OpenAI, Anthropic and Meta changed their usage policies in 2024 to allow U.S. intelligence and defense agencies to use their AI systems. Those policies still prohibit using the systems to harm humans.
Plumb said the Pentagon has been clear with AI model providers about what it will and will not use their technologies for. Even so, the article describes a rapid wave of partnerships between AI companies and defense contractors.
- Meta partnered with Lockheed Martin and Booz Allen to bring its Llama AI models to defense agencies in November.
- Anthropic teamed up with Palantir in November.
- OpenAI made a similar deal with Anduril in December.
- Cohere has also been deploying its models with Palantir.
These partnerships suggest that defense work is becoming a more accepted market for AI companies. They also raise a harder question: how far can military use go before it conflicts with a company's stated limits?
The answer is not clear from the outside. TechCrunch reported that it is unclear whose technology the Pentagon is using for work involving generative AI in the kill chain. The article also noted that even use at the early planning stage appears to create tension with the usage policies of several leading model developers.
The policy line is still hard to draw
Anthropic's policy, for example, prohibits use of its models to produce or modify “systems designed to cause harm to or loss of human life.” When TechCrunch asked Anthropic about the issue, the company pointed to a recent Financial Times interview with CEO Dario Amodei.
In that interview, Amodei argued against both extremes. He said the idea that AI should never be used in defense and intelligence settings does not make sense to him, while also rejecting the idea of unrestricted military AI development. He described Anthropic's position as an effort to find a responsible middle ground.
OpenAI, Meta and Cohere did not respond to TechCrunch's request for comment.
The policy challenge is that military support work is not always neatly separate from military action. A system used to organize information, model scenarios or suggest trade-offs may not pull a trigger. But it can still shape the chain of decisions that leads to force being used.
That is why the wording of company policies matters. It is also why the Pentagon's descriptions matter. If AI is framed as a planning tool, it may fit inside some corporate rules. If the same tool is understood as part of a process designed to prosecute threats, the boundary becomes more difficult to defend.
Humans in the loop remain the Pentagon's stated rule
The broader defense technology debate is not only about planning software. It also includes the question of whether AI weapons should be allowed to make life and death decisions.
Some argue that the U.S. military already has weapons that do. Anduril CEO Palmer Luckey wrote on X that the Department of Defense has purchased and used autonomous weapons systems for decades, including systems such as a CIWS turret. He said their use and export are understood, tightly defined and regulated by rules that are not voluntary.
Plumb pushed back when TechCrunch asked whether the Pentagon buys and operates fully autonomous weapons with no humans in the loop. “No, is the short answer,” she said. “As a matter of both reliability and ethics, we'll always have humans involved in the decision to employ force, and that includes for our weapon systems.”
Part of the dispute comes from the word “autonomy.” The article notes that the term is ambiguous and has fueled debate across technology fields, from AI coding agents to self-driving cars to self-firing weapons.
Plumb argued that the idea of machines independently making life and death decisions is too binary. She described the Pentagon's approach as human-machine teaming, with senior leaders making active decisions throughout the process.
Her point was that effective military AI is not a system that produces an answer for a person to rubber-stamp. It is a collaboration between people and machines, at least as the Pentagon presents it.
Why the debate is likely to grow
Military contracts have created backlash inside major technology companies before. TechCrunch noted that dozens of Amazon and Google employees were fired and arrested after protesting military contracts with Israel under the codename “Project Nimbus.”
By comparison, the response from the AI community has been more muted. Some researchers argue that military AI use is inevitable and that direct engagement with the government is necessary to reduce risk.
Anthropic researcher Evan Hubinger wrote in a November post on LessWrong that blocking the U.S. government from using AI is not a viable strategy if catastrophic AI risks are taken seriously. He also said that preventing misuse by the government remains necessary.
That captures the central tension. AI companies want to support government and defense customers without building weapons that make life and death decisions. The Pentagon wants faster, better tools for threat assessment and response. Between those goals sits the kill chain, where planning, targeting and force are connected parts of the same system.
For now, the public message from both sides is that humans remain responsible for force. But as generative AI becomes more useful to commanders, the pressure on policy language, company ethics rules and military oversight will only increase.