Autonomous Warfare's Open Questions: Control, Deterrence and Scale

Battlefields and budgets are pushing human judgment out of the kill chain, while diplomats keep affirming that it must stay in. Which force sets the norm is the central open question in autonomous warfare.

Key points

• Procurement tests in Ukraine are making machine terminal guidance standard faster than any treaty can define "human judgment."
• Diplomatic support for binding rules is growing, but the US, China and Russia sit outside the coalition or dilute its texts.
• Cheap interceptors make drone mass easier to defend against, yet deniable incursions and automated engagement create new escalation paths.
• The US is funding drones at an industrial scale while restricting the foreign supply chains that production depends on.
• The CCW Review Conference in November and the unpublished rewrite of DoD Directive 3000.09 are the nearest tests of where control is heading.

The tensions

Our signal scan found autonomy spreading faster than governance. Looking closer, the evidence does not all point one way.

Is human control eroding, or is a floor forming?

The case for erosion is strong. Fiber-optic drones have neutralized jamming, and each new countermeasure pushes both sides toward full onboard autonomy¹. Ukraine reports a tenfold rise in successful strikes with AI auto-homing since January, and six of seven vendors passed Ministry of Defence tests for procurement². Interception at the scale of 33,000 drones a month requires automated detect-and-engage loops³. China claims one operator can direct more than 200 drones under jamming, although state media is the only source⁴. In Washington, NSPM-11 ordered a deployment-first rewrite of the Pentagon's autonomy directive, which remains unpublished⁵.

The case for a floor rests on diplomacy. Seventy-six states now back negotiating a binding instrument⁶. Xi has publicly called for keeping AI "under human control," and Washington and Beijing are discussing an AI incident line⁷. Yet the major powers keep stepping back from commitments. Only about 35 of 85 states signed the REAIM declaration, and the US and China both declined⁸. The CCW rolling text was diluted after US and Russian pressure⁶.

What would settle it: whether the November Review Conference grants a negotiating mandate, and whether the revised 3000.09 keeps meaningful human judgment or narrows it to choosing a target before launch. We judge it likely, about 70%, that the working standard becomes target selection plus machine terminal guidance, whatever the treaty track produces.

Does cheap autonomous mass stabilize deterrence or strain it?

On the stabilizing side, defense is getting cheaper. Ukraine received twice as many interceptor drones in early 2026 as in all of 2025 and fields more than 25 domestic models³. Systems combat-tested in Ukraine are credited with more than 4,000 interceptions, and the Pentagon is buying them⁹. If cheap drones can reliably stop cheap drones, mass attack loses some coercive value.

On the destabilizing side, cheap mass lowers the cost of probing. About 20 Russian drones entered Polish airspace in 2025 under a deniable cover story, prompting Article 4 consultations and a standing NATO mission¹⁰. Automated border defenses raise the risk of misidentification¹⁰. Decision support adds another layer: frontier language models crossed the nuclear threshold in 95% of simulated crises, though the results may transfer poorly to real decisions¹¹.

What would settle it: incident data from Eastern Sentry and Poland's counter-drone network, and whether the US and China turn the incident line into a working channel with agreed red lines.

Can the US buy mass while decoupling its supply chain?

The money is real. The autonomous warfare group's request rose from $225.9M to as much as $54.6B⁹, and the Drone Dominance Program aims to buy about 340,000 small drones in two years¹².

Production is the constraint. By June, 20,000 FPV drones had been ordered, below the first phase target of 30,000¹². The FCC has moved from blacklisting DJI and Autel to proposing bans on whole categories of foreign drone hardware, while China's controls on rare-earth magnets limit Western supply in turn¹³. Russia still leads on fiber cable and spool volume¹. Each side holds a chokepoint it could use in a crisis.

What would settle it: whether later Drone Dominance phases meet their targets at fixed prices, and whether domestic motor and magnet supply grows fast enough to replace Chinese inputs.

Why these questions matter

While these questions stay open, track procurement test standards as closely as treaty texts, because norms on human control are being set in acceptance trials. Watch the November CCW decision, the 3000.09 rewrite and US drone order volumes. Planners who rely on a binding treaty or on uncontested US production scale should treat both as plausible rather than likely.