Pre-Launch Targeting Will Define Human Control by 2028

By 2028, human control over a drone strike will mean a person choosing the target before launch, with the machine finishing the attack, and no treaty the major drone powers sign will say otherwise.

Key points

• We put about 70% on pre-launch target selection being the working definition of "human judgment" in US and allied procurement by the end of 2028.
• The mechanism is already running: AI terminal homing passed Ukrainian acceptance tests, and the US is buying combat-proven systems with a request that rose from $226M to as much as $54.6B.
• The treaty track will likely split, about 60%, with a binding instrument negotiated outside the CCW that the US, Russia and China never join.
• Production, not policy, sets the pace: magnets, motors and cable decide how fast the standard spreads.
• The call fails if a fratricide or border misidentification forces a hard terminal-phase rule into law.

Our forecast

By the end of 2028, pre-launch target designation will be the operative definition of human control in US and allied drone procurement, written into acceptance criteria and accepted by the revised DoD Directive 3000.09 or its successor. We put this at about 70%. A binding instrument on autonomous weapons will likely exist by then, about 60%, negotiated outside the CCW and binding none of the US, Russia or China.

Why we think so

One pattern ran through the series: the standard is set by whatever passes a test and ships, and the rules follow.

The battlefield fixed the direction. Our signal scan showed fiber-optic tethers making jamming irrelevant at the front, with each countermeasure pushing both sides toward onboard autonomy as the next workaround¹. Brave1 reported a tenfold rise in successful AI auto-homing strikes between January and September, and six of seven vendors passed Ministry of Defence tests and are recommended for procurement². Defense is following: interceptors reportedly destroyed 33,000 Russian UAVs in a month, a volume only automated detect-and-engage loops can sustain³.

Procurement is importing that standard. Drone Dominance buys through fixed-price Gauntlet competitions⁴, and the autonomy group's request jumped to as much as $54.6B with $500M for systems combat-tested in Ukraine⁵. Whatever passed in Ukraine becomes the American requirement, the shadow norm our implications piece described.

Policy is moving to ratify, not restrain. NSPM-11 ordered a deployment-first rewrite of 3000.09 that admits commercial and open-source models, and no text has appeared four months later⁶. The plausible reading is that the rewrite legitimizes terminal autonomy already in service. Our contradictions piece asked whether diplomacy would form a floor. Not among the states that matter: the US and China both declined the REAIM declaration⁷, the CCW rolling text was diluted under US and Russian pressure, and consensus rules let the drone powers block the November mandate⁸. The 76 supporting states will likely take the treaty elsewhere, producing a norm with signatories but without fleets.

The convergences piece explains why the standard spreads at industrial rather than diplomatic pace. Supply chains are decoupling both ways, with US category bans on foreign drone hardware answered by Chinese magnet controls⁹, and the PLA's claimed 200-drone single-operator swarm, if real, has already moved past per-engagement control¹⁰. Neither bloc has an incentive to adopt a stricter standard than the other.

How this call fails

It is 2029 and we were wrong. Most likely, an incident forced a limit: an automated counter-drone system on NATO's eastern border shot down the wrong aircraft, or a US unit took fratricide from an auto-homing munition, and Congress used its 30-day notification right on 3000.09 changes to write a terminal-phase rule into the NDAA⁶¹¹. Second, Drone Dominance shortfalls persisted, the US fleet never reached the size that makes per-engagement control impractical, and the question stayed theoretical outside Ukraine⁴. Third, the US-China incident line matured into agreed red lines on autonomous engagement, and both found a floor cheaper than a race¹². We rate the incident path likeliest.

Who loses

• Middle powers and the treaty coalition, whose instrument will be real but govern none of the fleets that set the standard⁸⁷.
• Electronic-warfare incumbents and missile-interceptor makers, as tethers devalue jamming and cheap interceptors take the counter-drone mission¹³.
• Western drone producers dependent on Chinese magnets and motors, facing higher costs and slower output as bans bite from both sides⁹.
• Commanders on contested borders, who inherit automated engagement rules with no agreed trigger for answering deniable probes¹¹.

First moves

• Read the revised 3000.09 the day it appears and check one thing: whether "appropriate human judgment" is defined at launch or at engagement⁶.
• Treat Gauntlet scoring criteria and Brave1 test protocols as the real regulatory text, and track their changes as closely as treaty drafts²⁴.
• Audit any commercial model headed for decision support for escalation bias before adoption; the 95% nuclear-use result is a warning about defaults¹³.
• If you build drones or counter-drone systems, secure magnet, motor and cable supply outside China this year¹⁹.
• For arms-control work, plan now for a treaty outside the CCW, with verification that does not depend on the drone powers joining⁸.

What would change our mind

• A revised 3000.09 that requires human authorization of each engagement, not each launch, and survives the NDAA cycle⁶.
• The CCW Review Conference grants a negotiating mandate with US and Russian consent in November⁸.
• Later Drone Dominance phases hit their targets at fixed prices without Chinese inputs, which would remove our production caveat⁴⁹.
• The US-China incident line publishes red lines on autonomous engagement, a sign the two largest drone powers prefer a floor to a race¹².