The Trump administration is accelerating efforts to integrate artificial intelligence into U.S. military operations, even as some senior uniformed leaders and tech companies warn that safeguards are needed.
Admiral Frank Bradley, commander of U.S. Special Operations Command, told a special forces audience in Tampa that the services must be cautious as AI is introduced into systems that can deliver lethal force. He said he can imagine a future where AI recommends or selects targets, but stressed that humans must remain confident the systems will direct violence only where intended.
Bradley’s comments come amid a push from Defense Secretary Pete Hegseth to rapidly expand the Pentagon’s AI use. Hegseth has argued the military should be able to employ AI “in any legal way” and has rejected restrictions he says could limit wartime utility. President Donald Trump also halted plans for an AI executive order, saying he did not want to impede the U.S. lead in the technology.
Pentagon officials say current work emphasizes “functional battlefield tools” to help troops identify targets and accelerate decision-making. At the same time, Special Operations leaders describe many AI deployments as tools to free personnel from routine work and let operators focus on mission-critical tasks. Sgt. Maj. Andrew Krogman pointed to administrative relief and process modernization, while the command’s top acquisition official, Melissa Johnson, described AI as a means to lower cognitive load on mundane tasks and to augment—rather than replace—operator judgment.
Outside experts note both uses are real. Helen Toner, interim director of Georgetown’s Center for Security and Emerging Technology, said the military is exploring AI for bureaucratic efficiency as well as for combat support. Lt. Gen. Michael Conley of Air Force Special Operations Command told Congress his forces used AI “bots” to downgrade highly classified intelligence to a lower classification in seconds during the Iran conflict so drone crews could act on it more quickly.
There are also public examples of AI improving combat performance. A CSET case study found the Army’s 18th Airborne Corps used AI to direct artillery strikes at a level comparable to historically top-performing units while deploying about 2,000 fewer personnel. Analysts say AI enables operations at higher speed and scale, though humans still make the key decisions.
Debate over governance and control has played out in the open. The Pentagon’s relationship with AI firms has been strained. Anthropic’s CEO resisted allowing the company’s chatbot into classified Pentagon systems over concerns about how the technology could be used, including fears about fully autonomous weapons and intrusive surveillance. The department subsequently designated Anthropic a supply chain risk, ended a $200 million contract, and barred other contractors from working with the company; Anthropic sued, alleging unlawful retaliation.
In response, the Pentagon has turned to other vendors, including Google, OpenAI and SpaceX, to procure AI capabilities intended to support warfighter decision-making in complex environments.
Observers note the public friction masks a deeper reality: the U.S. military often approaches new technologies cautiously. Commanders want capabilities that can create decisive effects, but they also must avoid unintended consequences such as friendly fire, civilian harm, or target misidentification. That tension—between speed, capability and ethical constraints—continues to shape how AI is adopted, who controls it, and what rules govern its battlefield use.