In February 2026, a confrontation between Anthropic and the Pentagon became public. Anthropic’s Claude model had been integrated into the Maven Smart System via Palantir, and when the Department of Defense sought to deploy it in fully autonomous lethal weapons systems without human oversight, Anthropic refused. Defense Secretary Pete Hegseth argued that the Pentagon couldn’t be constrained by a vendor’s internal safety policies. Anthropic countered that specific defensive scenarios—like laser interception of incoming drones—could be addressed case by case without abandoning a general prohibition on AI-powered lethal autonomy. The Pentagon rejected that approach as operationally unworkable.
That exchange crystallized the question that has been building for a decade and is now unavoidable: when combat happens faster than a human can think, who decides whether the machine is allowed to kill?
The speed problem
The theoretical debate about human-in-the-loop weapons systems is increasingly being resolved by physics. A drone swarm approaching a military installation at speed, a hypersonic missile in its terminal phase, an adversary FPV drone inside the perimeter of an air base—all present threat timescales measured in seconds. A human being requires roughly 200 to 300 milliseconds to perceive a stimulus and initiate a motor response under ideal conditions. Add situational assessment, rules-of-engagement evaluation, and communication latency, and the realistic human response time for a complex targeting decision is measured in seconds to minutes. The engagement window for many modern threats is shorter than that.
This is the argument the Pentagon made to Anthropic. Israel’s Iron Beam laser system, whose accelerated deployment began in late 2025, uses autonomous targeting to neutralize incoming threats at speeds no human operator could match. The system identifies, tracks, and engages projectiles in a fraction of the time a human decision-maker would need to process the same information. South Korea has deployed autonomous sentry systems along the DMZ since the mid-2000s—the Samsung SGR-A1, equipped with machine guns and pattern-recognition software, can detect and track intruders and theoretically fire without human authorization, though human approval is currently required. Russia announced serial production of the Marker ground combat robot in March 2025, equipped with Kornet anti-tank missiles and drone swarm coordination capabilities.
In December 2025, Auterion demonstrated the first multi-manufacturer combat drone swarm—a single operator directing FPV platforms and fixed-wing loitering munitions from different manufacturers as a coordinated force. The demonstration pointed toward a future where one human supervises dozens or hundreds of autonomous lethal platforms simultaneously, which redefines “human in the loop” to something closer to “human vaguely aware of the loop.”
What DoD Directive 3000.09 actually says
The U.S. policy framework for autonomous weapons is DoD Directive 3000.09, originally issued in 2012 and updated in January 2023. The directive does not ban autonomous weapons. It establishes a review and approval process for their development and deployment, and it requires that autonomous and semi-autonomous weapons systems be “designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.”
The key phrase is “appropriate levels.” The directive doesn’t define a single standard for human control. It creates a spectrum—from fully human-controlled systems to systems with increasing degrees of autonomy—and requires that the level of human involvement be calibrated to the operational context. A defensive system intercepting an incoming missile has different human-oversight requirements than an offensive system selecting and engaging a human target. The 2023 update explicitly acknowledged AI-enabled systems and reinforced that existing international humanitarian law—the principles of distinction, proportionality, and military necessity—applies to autonomous weapons regardless of the degree of automation.
In practice, this means the United States has positioned itself to develop and deploy autonomous weapons systems across the full spectrum of autonomy while maintaining that human judgment is preserved at “appropriate” points in the kill chain. Critics argue that “appropriate” is doing an enormous amount of work in that sentence, and that the directive’s flexibility is a feature, not a bug—it enables autonomous weapons development without the political cost of explicitly authorizing machines to kill without human approval.
The FY2026 numbers
The Pentagon requested a record $14.2 billion for AI and autonomous systems research in its fiscal year 2026 budget. The Replicator program—designed to fast-track deployment of thousands of expendable autonomous drones and surface vessels—received $1 billion in 2025. These are not research abstractions. They are procurement programs producing hardware that is entering or about to enter operational service.
The fiscal year 2026 NDAA, which authorized $900.6 billion in defense spending, contains provisions addressing autonomous weapons within the broader framework of emerging technology governance. The legislation requires reporting on AI-enabled systems but does not impose new restrictions on autonomous weapons development or deployment beyond those in Directive 3000.09.
Ukraine has served as the world’s largest testing ground for autonomous weapons concepts. Both sides have deployed increasingly autonomous drones—FPV kamikaze drones with AI-assisted targeting, loitering munitions with pattern recognition, and experimental ground robots. The conflict has demonstrated that low-cost autonomous systems can be decisive in modern warfare, and that the nations watching the conflict—which is all of them—are incorporating those lessons into their own procurement and doctrine at extraordinary speed.
The treaty that isn’t happening
The UN Secretary-General called for a legally binding treaty prohibiting lethal autonomous weapons systems from operating without human control, with a 2026 target completion date. The UN General Assembly passed Resolution 79/62 in December 2024 with 166 votes in favor, mandating informal consultations among member states. The Group of Governmental Experts on LAWS, operating under the Convention on Certain Conventional Weapons, has been discussing the issue since 2014.
The treaty is not happening. Not by 2026, and likely not in any meaningful form while the three largest military AI developers—the United States, Russia, and China—oppose binding restrictions. The U.S. and Russia voted against the November 2025 resolution calling for negotiation of a legally enforceable LAWS agreement. China has expressed support for regulation in principle while continuing to develop autonomous weapons capabilities. Israel, whose national defense increasingly depends on autonomous interception systems, also voted against.
The Arms Control Association described the current moment as the “pre-proliferation window”—the last opportunity to establish norms before autonomous weapons become as widespread and unmanageable as small arms. The Stop Killer Robots coalition, led by a network of civil society organizations, advocates a two-tiered approach: an outright prohibition on fully autonomous weapons that target humans, combined with regulation of systems with greater human control. The International Committee of the Red Cross has called for legally binding rules by 2026.
The problem is structural, not rhetorical. The nations with the most advanced autonomous weapons programs are precisely the nations that would need to agree to restrictions for those restrictions to matter, and they have no strategic incentive to constrain capabilities they’ve invested billions in developing. The CCW operates by consensus, meaning any single state can block progress. The Group of Governmental Experts’ mandate extends to 2026, with the CCW Review Conference set as the deadline for a final report, but the likelihood of that report containing binding restrictions that the U.S., Russia, or China would accept approaches zero.
The accountability gap
International humanitarian law requires that someone be held responsible for every use of lethal force. The principle of distinction requires distinguishing combatants from civilians. The principle of proportionality requires that civilian harm be proportional to the military advantage gained. The principle of military necessity requires that force serve a legitimate military objective. These principles were designed for a world in which a human being makes the decision to fire.
When an algorithm makes that decision—or when the decision happens at machine speed with nominal human oversight—the accountability framework fractures. If an autonomous weapon misidentifies a civilian as a combatant and kills them, who is responsible? The manufacturer who designed the targeting algorithm? The commander who authorized deployment? The software engineer who trained the model? The procurement officer who selected the system? The question isn’t hypothetical. In Gaza, reporting has described AI-generated target lists of tens of thousands of individuals, with automated systems recommending strikes at a pace that effectively eliminates meaningful human review of individual targeting decisions.
The ethical objection—articulated by the UN Secretary-General as “morally repugnant”—is that delegating the decision to take a human life to a machine violates human dignity regardless of how accurately the machine performs. The pragmatic counterargument—articulated by every major military power investing in autonomous weapons—is that the alternative is slower, less accurate human decision-making that results in more casualties, including more civilian casualties, because humans under stress make worse targeting decisions than well-designed algorithms.
Both arguments are sincere. Both arguments are partially correct. And the tension between them is not going to be resolved by a treaty, because the nations building these weapons have decided that the strategic advantages outweigh the ethical costs, and no international legal framework has ever successfully constrained a weapons technology that major powers considered essential to their security.
The kill chain is getting shorter. The human role within it is getting thinner. The governance framework is getting further behind. Whether that trajectory is inevitable or merely current policy is the question that 2026’s diplomatic calendar—the CCW Review Conference, the GGE final report, the continuing fallout from the Anthropic-Pentagon confrontation—will begin to answer, even if nobody expects the answer to be satisfying.
We cover autonomous weapons, drone warfare, electronic warfare, and the full spectrum of emerging military technology across 24 lectures in our Battlefields of the Future course—including why the nations building these weapons and the nations trying to ban them are operating on fundamentally incompatible timelines.