The U.S. military is accelerating the integration of AI into its combat systems, but senior military officials remain cautious about the boundaries of its use in lethal strikes. General Frank Bradley, commander of U.S. Special Operations Command, recently stated that AI may be involved in identifying targets in the future, but humans must ensure that force only lands where intended.
Military leaders emphasize human oversight
Speaking at a special forces annual conference in Tampa, Florida, Bradley said that forces must be extremely cautious when using AI in lethal operations. He suggested that in the future, AI might be able to decide which targets to strike, but a key prerequisite is that operators must have sufficient confidence in the system.
This statement comes as U.S. Defense Secretary Peter Hegseth is pushing for the military to accelerate the adoption of AI. Hegseth previously stated that the Pentagon should be able to use the technology at its own discretion, provided it complies with legal requirements. In January of this year, he also stated that he would not accept AI models that were "unusable for combat."
The U.S. military has given conflicting descriptions of its internal use.
Pentagon officials stated that the current focus is on making AI a usable tool on the battlefield, helping troops identify and target targets more quickly, thereby shortening strike time. Meanwhile, the U.S. Special Operations Command has publicly emphasized that AI is primarily an auxiliary tool to reduce personnel workload and improve mission efficiency.
Senior Sergeant Andrew Krogman of the command stated that AI can handle administrative tasks, allowing warfighters to focus more on the mission itself. Melissa Johnson, who is in charge of procurement, also said that the role of AI should be to reduce the cognitive burden of repetitive tasks, rather than to replace warfighters' judgment.
AI has entered the intelligence and strike process.
Helen Toner, acting executive director of Georgetown University's Center for Security and Emerging Technologies, believes these two statements are not contradictory. She stated that while the U.S. military is indeed extensively exploring the use of AI in administrative and procedural processes, AI has also been used to more quickly detect and engage targets.
U.S. Air Force Special Operations Command Commander Michael Conley told Congress this month that his forces used AI robots during the Iran war to convert top-secret intelligence into lower-class information within seconds for faster sharing with ground drone operators. A previous case study by Toner's organization also mentioned that the U.S. military used AI to improve artillery target selection efficiency while reducing personnel deployment by approximately 2,000.
Pentagon in open conflict with Anthropic
The conflict between the Trump administration and Anthropic over the security boundaries of military AI has become public. Key points of contention include whether the government will use the technology without sufficient restrictions, and the risks of fully autonomous armed drones and AI-assisted mass surveillance.
The report states that after Anthropic CEO Dario Amodei refused to ease concerns about Claude's use of classified networks within the Pentagon, both Trump and Hegseth accused the company of jeopardizing national security. Subsequently, the Pentagon designated Anthropic a supply chain risk, terminated its $200 million defense contract, and banned other government contractors from working with it.
Additional information:Subsequently, the Pentagon more explicitly turned to companies such as Google, OpenAI, and SpaceX, seeking AI technologies that could be used to improve the efficiency of operational decision-making; Anthropic has filed a lawsuit regarding the relevant handling.
