Meaningful Human Control as Architecturally-Defined

Where the Doctrine Stands in 2026

The CCW GGE on LAWS, operating under its rolling text since 2023 and renewed mandates through 2026, has converged on a two-tier framing: prohibitions on systems that operate beyond meaningful human control, and regulations on systems that operate within it. The text remains contested at the prohibition tier, but the regulation tier has crystallized around MHC as the necessary condition. The ICRC's 2021 position, reinforced in subsequent expert meetings, holds that human control must be exercised over the use of force in a manner sufficient to ensure compliance with international humanitarian law; this is widely treated as the floor that any GGE outcome must clear.

National declarations have populated the space the GGE has not yet closed. France and Germany's joint working papers, the Brazilian and Austrian sponsorships of stronger control language, the Netherlands' AREA framework, the United Kingdom's Defence AI Strategy and JSP 936 doctrine, Australia's Method of Warfare doctrine update, and U.S. Department of Defense Directive 3000.09 (revised 2023, with implementation through CJCS instructions in 2024 and 2025) all push toward operationally-specified MHC. The U.S. directive is notable for its explicit requirement that autonomous and semi-autonomous weapon systems allow commanders and operators to exercise appropriate levels of human judgment over the use of force — and for its 2023 revision tightening the senior-review pathway for systems that approach the autonomous-engagement boundary.

Article 36 of Additional Protocol I to the Geneva Conventions binds states to review new weapons, means, and methods of warfare for compliance with international humanitarian law. The convergent reading among ICRC, GGE participants, and national legal advisors is that Article 36 reviews of autonomous systems must address MHC structurally, not merely descriptively. This shifts the burden onto the architecture: a weapon system whose architecture cannot expose how human control was exercised, against which target classes, with what authority, and reconcilable to which post-action lineage, will increasingly fail Article 36 review in the states whose review processes are converging on the GGE direction of travel.

What 'Meaningful' Structurally Requires

'Meaningful' is the operationally-pivotal qualifier in MHC, and the academic literature (Article 36 the NGO, the ICRAC papers, the SIPRI work, the Roff/Moyes framework) has converged on a roughly consistent set of structural conditions. Operator intent must be predicate to engagement: the operator must have formed and expressed an intent that the engagement falls within, and that intent must be auditable. The system must operate within a bounded operational envelope — spatial, temporal, target-class — that the operator's intent specifies, and the system must structurally refuse engagements outside the envelope. The operator must retain the ability to intervene on a timescale meaningful to the engagement, and the architecture must support that intervention as a first-class operation rather than as a best-effort interrupt.

Translated to engineering, MHC requires that operator intent is architecturally-recorded rather than reconstructed from logs after the fact, that engagement actuations admit structurally against the recorded intent rather than being filtered post-hoc against policy, that intent transitions (operator takeover, intent escalation, intent revocation, command handover between operators or between fleets) enter the same lineage as the actuations they govern, and that post-action audit can verify intent-to-actuation correspondence end-to-end without relying on operator self-report or after-the-fact reconstruction.

Each of these requirements maps to an architectural primitive that operator-intent exposes. Without architectural support, MHC operates as policy intention overlaid on systems that do not natively distinguish between an authorized engagement and an unauthorized one until the action has already been taken. With architectural support, MHC operates as a structural property of the system: an engagement that does not admit against active credentialed intent does not occur, and an engagement that does occur carries its lineage to the audit.

The Additional Protocol I obligations under Articles 48 (distinction), 51 (protection of civilians), and 57 (precautions in attack) sharpen the structural demand. Distinction requires that the system act only against lawful targets; this is testable structurally if intent declaration scopes target classes and admissibility refuses out-of-scope engagement. Article 51's proportionality and Article 57's precautions are testable structurally if the architecture can produce, for each engagement, a lineage showing what the operator intended, what alternatives were available, and what verification was performed before commitment. None of this is achievable post-hoc against a system that did not architecturally record the answers in the first place.

How Operator-Intent Maps to MHC

Credentialed intent declaration provides the architectural intent recording that MHC requires. The operator (or chain of authority delegating to the operator) declares intent that is cryptographically bound to identity, scope, and operational envelope. The declaration is a structural artifact, not a free-text annotation, and the system's admissibility logic operates against it directly. Multi-fleet, multi-authority intent extends the primitive to the realistic case where engagement authority spans coalition partners, joint task forces, or cross-domain operations under different national rules of engagement.

Composite admissibility against active intent provides structural admission of every engagement. The actuation does not commit unless the candidate engagement admits against the active intent under all credentialed authorities — legal review, command authority, engagement-rules enforcement, target-class constraint. Refusal is the default and admission is the structurally-evidenced exception. This is the inversion that distinguishes architecturally-supported MHC from policy-overlay MHC: the system structurally cannot engage outside the authorized envelope, rather than being policy-discouraged from doing so.

Intent-transition recording provides lineage across the dynamic moments where MHC most often fails in practice. Operator takeover during an autonomous engagement, escalation of intent in response to a developing situation, revocation of intent when the situation no longer warrants engagement, handover between operators or between fleets — each of these is a transition that, in current architectures, is typically reconstructable only from disparate logs. Architectural intent-transition recording binds the transition into the same lineage as the actuations that occur on either side of it, so the audit can answer the question 'who held intent at the moment this actuation committed' without ambiguity.

Audit traversal across intent-actuation provides the verification surface that Article 36 reviews and post-incident investigations both increasingly demand. Graduated fidelity tiers let the audit operate at the level of fidelity the inquiry requires — high-fidelity per-actuation lineage for incident review, lower-fidelity aggregate for routine compliance reporting — without re-deriving the artifacts. The regulator-as-credentialed-observer pattern lets a treaty-monitoring authority, a national legal advisor, or an independent investigator subscribe to the lineage under specified conditions without requiring the operator to curate an extract that itself becomes a point of trust.

MHC, under this mapping, becomes operationally-real rather than policy-aspirational. The system either admits engagements against credentialed intent or it does not; the lineage is either traversable or it is not; the transitions are either structurally recorded or they are reconstructed. The doctrine's qualifier, 'meaningful', acquires an engineering definition.

Where National and International Doctrine Are Heading

The GGE rolling text is converging on language that requires structural demonstration of MHC for systems above a threshold of autonomy in the use of force. The threshold remains contested, but the demonstration requirement does not. National doctrines are populating the space below the threshold with their own structural expectations: UK MoD's emerging LAWS policy, the Australian Department of Defence's autonomous-systems policy under the Method of Warfare doctrine, the Netherlands AREA framework, France's 2021 autonomous-weapons policy paper, and U.S. DoD Directive 3000.09 (2023 revision) all increasingly require that MHC be evidenced architecturally rather than asserted procedurally.

Article 36 review processes in states with mature legal-review capacity (UK, Netherlands, Sweden, Norway, Australia, Canada, New Zealand, increasingly Germany and France) are tightening. Reviewers no longer accept that a system 'has a human in the loop'; they ask what the human's role structurally is, what the system structurally does when the human's input is delayed or absent, what the lineage of an engagement structurally consists of, and how the system would behave under the failure modes of the human-machine interface itself. These are architectural questions. They are answered structurally or they are answered insufficiently.

The trajectory among non-state actors and academic-civil-society coalitions reinforces the direction. Article 36, ICRAC, SIPRI, the Stop Killer Robots coalition, and the ICRC-sponsored expert meetings continue to push the structural reading of MHC into formal record at GGE sessions and parallel forums. Even where states diverge on the prohibition tier, the regulation tier's structural-MHC requirement is being baked into the procurement and operational-test expectations the major defense ministries impose on their suppliers.

The credentialed-observer pattern is the next-frontier expectation. Treaty-monitoring proposals at the GGE, the discussion of confidence-building measures around autonomous systems, and the bilateral and minilateral arrangements emerging around responsible AI in military domains all point toward an architecture where authorized external observers can subscribe to lineage under specified conditions. An architecture that exposes the credentialed-observer role natively is positioned for that frontier; an architecture that requires curated extraction and operator filtering is not.

Position

Meaningful Human Control will not be defined by a single instrument; it will be progressively specified by the GGE rolling text, by Article 36 review practice in states with mature review capacity, by national doctrine updates, and by the operational-test expectations that flow into procurement. The direction of travel converges on structural demonstration: credentialed intent, composite admissibility, lineage, and audit traversal, with credentialed-observer access as the emerging frontier expectation. Operator-intent supplies the architectural substrate at exactly that layer. Operators of autonomous and semi-autonomous defense systems that internalize the substrate now produce MHC evidence by construction across every engagement, every doctrine update, and every Article 36 cycle; operators that defer continue to assert MHC procedurally against doctrinal review processes that are increasingly unwilling to accept procedural assertion as sufficient.