In September 1983 a Soviet early-warning system reported five incoming American missiles. The officer on duty, Stanislav Petrov, decided the computer was wrong and did not pass the alert up the chain. He was right. Forty years later, militaries are building weapons that find, track, and strike targets with less and less room for a Petrov, and the law of war has to decide what "meaningful human control" means when the human is a thousand miles away and the machine is faster than thought.
In this TalksOnLaw interview, recorded in November 2024, Joel Cohen speaks with Mitt Regan, the McDevitt Professor of Jurisprudence at Georgetown Law and Director of its Center on Ethics and the Legal Profession, who has spent the past decade studying the ethics and law of autonomous weapons, about what these systems actually do, where international humanitarian law bites, and why the treaty negotiations have stalled.
What Autonomy Actually Means
Regan dismantles the image of a "killer robot" deciding on its own. AI already runs throughout the targeting cycle, from sorting intelligence to recommending targets to guiding munitions, and the real question is which functions are delegated and under what constraints. He explains "in the loop" versus "on the loop," the automation bias that leads operators to defer to a confident screen, the Ukrainian and Russian loitering drones that can complete a strike after losing their link, and why ship-defense systems like Aegis, which fire automatically at incoming missiles, have been accepted for decades: "the notion of human control is a property of the system as a whole," not of whoever presses the last button.
The Law of War Meets the Algorithm
He works through distinction, proportionality, and precautions in attack, the Article 36 weapons-review obligation, and the ICRC's position that systems should never select human targets. He explains why explainability matters legally (a commander must be able to justify a strike), how adversaries can spoof a targeting model, and what the reporting on Israel's AI-assisted target generation in Gaza suggests about the pace and volume of machine-recommended strikes. He also previews his current work on the "paradox of precision": weapons that hit exactly what they aim at while the cumulative destruction of homes, hospitals, water, and power causes most of the civilian harm, a form of damage the law has barely begun to account for.
Treaties, Politics, and the Road Ahead
Regan describes the decade of deadlock in the Convention on Certain Conventional Weapons, where consensus rules let any major military power block a binding instrument, the alternative of a UN General Assembly track, and the reasons the United States prefers policy to treaty. "It's nuanced all the way through," he says, and he makes the case that lawyers who understand the technology have a role to play in shaping the constraints before the constraints are set by events.
What to Know Now
The diplomatic picture has moved quickly. In December 2025 the UN General Assembly adopted Resolution 80/57 on lethal autonomous weapons by a vote of 164 to 6, and in September 2026 the CCW Group of Governmental Experts finally reached consensus on a set of "elements" for a possible instrument, setting up the CCW Review Conference in November 2026 to decide whether to open formal negotiations. The UN Secretary-General and the ICRC renewed their joint call for a binding treaty by the end of 2026. On the U.S. side, the Pentagon's Replicator initiative was folded into a new Defense Autonomous Warfare Group, the FY2027 budget requested $54.6 billion for autonomy and drone programs, and Department of Defense Directive 3000.09 on autonomy in weapons is under revision following a June 2026 national security presidential memorandum. The 2026 conflict with Iran produced the first widely reported case of a strike attributed in part to over-reliance on an AI targeting tool, at the port of Minab, and Microsoft's September 2025 decision to cut an Israeli military intelligence unit's access to its cloud services showed that the technology companies supplying these systems have become actors in the debate.
