An era is approaching in which artificial intelligence (AI), not humans, chooses whom to shoot on the battlefield. Under the Trump administration, the U.S. military is elevating AI from an assistant tool that summarizes documents and helps with administration to an auxiliary brain for the kill chain that finds and strikes the enemy. The kill chain refers to the series of processes of detecting, identifying, and striking the enemy. Within the military leadership and parts of the tech industry, a cautious view has begun to emerge that humans could be reduced to merely stamping forms.
On the 31st (local time), according to a compilation of major foreign outlets including the AP, U.S. Special Operations Command chief Frank Bradley, envisioning a future in which AI sets targets, called for strict control measures at a recent annual special operations conference in Tampa, Florida. Bradley said, "We must be very cautious about how we employ AI and how we leverage it for lethality," adding, "As humans, we need to be confident that AI will apply violence only where we intend."
Bradley is the head of special operations who oversees the most demanding and covert missions in the U.S. military. He did not oppose bringing AI onto the battlefield itself, but he expressed vigilance, saying a "much higher level of caution is necessary." The comment is read as a warning that the moment technology is combined with lethal systems, it can produce unintended results.
The AI the Pentagon wants is not at the level of a chatbot. The battlefield is a continuum of finding where the enemy is, verifying whether it is a real target, deciding which weapon to use, obtaining higher approval, and executing. The U.S. Department of Defense seeks to deploy AI to compress this entire process. The logic is that the side that sees first, decides first, and strikes first wins.
According to major military outlets, the U.S. military now parses drone footage, reconnaissance satellite imagery, radar signals, and signals intelligence all at once to shortlist candidate targets for strikes in real time. In the past, analysts replayed footage frame by frame to find armories and unit movements. Now algorithms read pixel changes and heat signatures on the screen to automatically identify potential threats. The U.S. Department of Defense views boosting this decision speed with AI as a key means to overwhelm adversaries in decision-making dominance.
According to a case study by the Center for Security and Emerging Technology (CSET) at Georgetown University, advanced AI is not a far-off technology but is already used in some operations. The U.S. Army's XVIII Airborne Corps succeeded in performing target processing work—once backed by more than 2,000 personnel—with a small unit of about 20 people by leveraging AI. This is the result of handing over the manual flow of finding targets, verification, approval, weapons assignment, and damage assessment to AI. Helen Toner, acting director at CSET, said, "Humans still make the decisive calls, but AI enables operations at a new level of speed and scale." U.S. Air Force Special Operations Commander Michael Conley said at a congressional hearing last month that unit members used AI "bots" to downgrade top secret information to secret in seconds and share it in real time with drone operators in the Iran theater.
Those urging caution about introducing AI to the battlefield worry about the moment the same technology crosses into target selection and strike decisions. The U.S. military's "autonomous weapons" guidelines stipulate that "autonomous and semi-autonomous weapons systems must be designed to allow commanders and operators to exercise appropriate levels of human judgment in the use of force." However, if AI takes on the core of the kill chain—target identification and prioritization—it is difficult to fully uphold this principle. On the battlefield, key decision factors often exceed human cognitive capacity. In situations where life and death are decided in seconds, commanders do not have enough time to validate AI analyses one by one. If they accept as is a system in which AI presents dozens of candidate targets and even recommends risk scores and suitable strike weapons, on the surface humans bear final responsibility, but in substance the decision-making shifts to AI algorithms. Caution advocates worry this trend could ultimately spill over into fully autonomous armed drones that find and attack targets without human intervention and into AI mass-surveillance systems.
Missy Cummings, who leads the Robotics and Automation Center at George Mason University, said there is a constant risk that AI could mistake civilian vehicles for enemy transport trucks or be fooled by jamming and fake data into striking the wrong place. In a paper published in Dec. last year, Cummings said, "Government agencies should prohibit the command, control, and operation of weapons with Generative AI," adding, "The large language models that underpin AI chatbots make too many errors—such as hallucinations and fabrications—to be trusted. They will kill noncombatants and they will kill friendly forces."