Trump’s AI mandate: Defining killer robots won’t save us — hardcoding human control will

President Donald Trump gave the Pentagon 90 days to update the rule governing autonomy in weapon systems. That window closes this week. The administration’s instruction was not simply to adopt artificial intelligence faster. It was to do so while preserving the chain of command and operational authority.That gives the Pentagon an opportunity to solve a harder problem than definition alone can reach.One influential proposal in the current debate, advanced by the Center for Strategic and International Studies, is to broaden the definition of an autonomous weapon so that it reaches beyond drones and munitions to the software that fuses sensor feeds, selects targets, and orchestrates the kill chain. That makes sense. If software performs the function that makes a weapon autonomous, policy should not stop at the airframe. But a wider definition is only the first step. A definition tells the military what falls under a rule. It does not tell commanders what authority humans must retain once a system falls inside it.The current Directive 3000.09 already requires autonomous and semiautonomous weapon systems to allow commanders and operators to exercise “appropriate levels of human judgment” over the use of force. It also requires rigorous testing, transparent system status, and systems that terminate an engagement or seek additional operator input when they cannot continue within the commander’s intended constraints.Those are strong foundations. The update can make them more operational in three ways.First, it should specify human authority, not merely the amount of human involvement. For a given system and mission, the rules should identify which decisions remain reserved to a person, under what conditions the system may act without additional input, and which boundaries it may never cross on its own. Call this an authority envelope: a mission-specific statement of what the machine is permitted to do and what remains a human decision.Second, that boundary should be enforced before action. If a decision is reserved for a human, the system should be technically unable to cross that line until the required authorization has occurred. A safeguard discovered only in an after-action review was not a safeguard at the moment it mattered.Third, human control should be verifiable, not merely asserted. A system should preserve enough of the decision record to establish what information was presented, what the machine recommended, what the human authorized, what operational limits were active, and in what sequence those events occurred. The goal is not to automate legal judgment. It is to preserve evidence that the human role required by policy actually existed in practice.None of this asks the Pentagon to slow AI adoption. It builds directly on principles the department has already embraced. The War Department’s AI ethics framework calls for systems to be traceable, reliable, and governable, including the ability to disengage or deactivate systems that behave in unintended ways. The new autonomy rule can translate those principles into testable requirements for systems that can affect the use of force.That matters because machine speed changes what “human in the loop” can mean. A person can remain formally present while the decision window becomes too short to understand the evidence, challenge the recommendation, or exercise a reserved authority before the action is complete. The answer is not to demand a human click at every millisecond. It is to decide in advance which decisions must remain human and to engineer the system so those decisions cannot silently migrate to software as tempo increases.Trump’s memorandum set the right objective: accelerate military AI while preserving clear accountability and the chain of command. The Pentagon now has a chance to make that objective durable beyond any one administration or technology cycle.BIG TECH’S AI FUTURE RELIES ON AN UNLIKELY HERO: THE BLUE-COLLAR WORKERA successful update to Directive 3000.09 would do more than define autonomous weapons more accurately. It would tell commanders what decisions remain theirs, engineers what boundaries must be enforced, and testers what evidence must exist to show that human authority survived machine speed.The point is not simply to define the autonomous weapon. It is to make human command technically real.Burak Oktenli is an independent researcher based in Washington, D.C. He holds a bachelor’s degree in computer science and engineering from the University of South Florida and an MBA, and is completing a master’s of professional studies in applied intelligence at Georgetown University. His writing focuses on defense strategy, alliance burden-sharing, military technology, and the governance of autonomous systems.