Department of Defense Directive 3000.09, "Autonomy in Weapon Systems," is the U.S. policy governing the design, development, acquisition, testing, fielding, and use of autonomous and semi-autonomous weapon systems across all DoD Components. It establishes a design standard built around "appropriate levels of human judgment over the use of force," a set of design requirements, and a senior-review mechanism for higher-risk systems. The current text is a 2023 reissuance of the original 2012 directive. (Source: source summary)
Status and timeline
The directive was issued by the Office of the Under Secretary of Defense for Policy and signed by Deputy Secretary of Defense Kathleen H. Hicks, taking effect January 25, 2023. It reissued and cancelled the prior DoD Directive 3000.09 of November 21, 2012 (incorporating Change 1 of May 8, 2017), which had originally been issued by Deputy Secretary Ashton Carter. Under the 10-year reissuance cycle set by DoD Directive 5025.01, the directive is due for reissuance, cancellation, or certification of currency no later than January 25, 2033.
On May 22, 2026, Defense Department officials told lawmakers they plan to update the Pentagon's policy on human control of autonomous weapons systems, meaning Directive 3000.09 itself. Officials noted that the policy had not been revised in more than three years (the current text took effect January 25, 2023) while the department's adoption of autonomy technologies had accelerated. This would be the first revision since the 2023 reissuance and arrives well inside the nominal 10-year reissuance cycle. The scope and direction of the planned changes, including whether they would tighten or further soften the "appropriate levels of human judgment" standard, the senior-review threshold, or the AI-specific provisions, were not specified. (Source: insideaipolicy.com)
Scope and definitions
The directive applies to the design, development, acquisition, testing, fielding, and employment of autonomous and semi-autonomous weapon systems, including guided munitions capable of automated target selection, and to the application of lethal or non-lethal, kinetic or non-kinetic force by such systems, across all DoD Components. It excludes cyberspace operations systems; unarmed, unmanned platforms; unguided munitions; manually guided munitions (laser- or wire-guided); mines; unexploded ordnance; and research and development activities intended to improve the safety or effectiveness of such systems.
Three categories of system are defined. An autonomous weapon system, once activated, selects and engages targets without further human intervention. A semi-autonomous weapon system engages only individual targets or specific target groups selected by a human operator; this category includes fire-and-forget and lock-on-after-launch homing munitions, provided a human selected the target. A human-supervised autonomous weapon system is autonomous but designed so that human operators can intervene and terminate engagements.
Core policy and design requirements
The directive states that autonomous and semi-autonomous weapon systems will be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force. This 2023 formulation replaced the 2012 wording by dropping the word "control," a change criticized by Human Rights Watch (see Reactions).
Designs must incorporate rigorous verification and validation and realistic developmental and operational test and evaluation, including testing for unanticipated emergent behavior; the capability to complete engagements within a timeframe, geographic area, and environmental constraints consistent with commander and operator intentions, or else terminate engagements or seek operator input; sufficient robustness against failures that could lead to unintended engagements or loss of control; operator-facing design providing the means to activate and deactivate the system, clear procedures, and training so operators understand the system's capabilities and limits; and consistency with DoD's Responsible AI Strategy and the five DoD AI Ethical Principles (Responsible, Equitable, Traceable, Reliable, and Governable).
The 2023 reissuance added provisions specific to AI-using autonomous and semi-autonomous weapon systems. Such systems must be transparent, auditable, and explainable to relevant personnel to the maximum practicable extent; use auditable data, methods, processes, models, and measures; be capable of rapid reprogramming on new input data; undergo test and evaluation sufficient to verify AI robustness, effectiveness, and suitability; and provide the capability to disengage or deactivate deployed systems exhibiting unintended behavior.
Senior review
The directive's central control mechanism is a senior review required both before formal development and again before fielding for certain systems. The required approvers are the Under Secretary of Defense for Policy (USD(P), the lead), the Under Secretary of Defense for Research and Engineering (USD(R&E)), and the Vice Chairman of the Joint Chiefs of Staff (VCJCS).
Systems that trigger senior review include autonomous weapon systems intended for use outside policy guidelines; human-supervised autonomous systems used to select and engage targets (subject to a limited local-defense exception for time-critical or saturation attacks on manned installations and platforms); and substantially modified previously reviewed systems, where changes to algorithms, mission sets, environments, or target sets "substantially differ." The Autonomous Weapon Systems Working Group, created by the 2023 update, assists with submissions, which are routed directly to USD(P). The Deputy Secretary of Defense may waive senior review in cases of urgent military need.
Per DoD's own public acknowledgment, no weapon system has been required to undergo this senior review since the directive's original 2012 issuance, a point central to the Human Rights Watch critique below.
Responsibilities
The directive assigns roles across DoD. USD(P) is the policy lead, maintains the Autonomous Weapon Systems Working Group, and receives senior-review submissions. USD(R&E) sets technical and test-and-evaluation standards and participates in senior review. The Under Secretary of Defense for Acquisition and Sustainment (USD(A&S)) integrates acquisition policy. The Chief Digital and AI Officer (CDAO) aligns the directive with the DoD Responsible AI Strategy and AI Ethical Principles. VCJCS supplies the joint warfighting perspective and participates in senior review. DoD Component Heads are responsible for compliance, test and evaluation, and reporting. The DoD General Counsel conducts legal review of weapons per DoD Directive 5000.01 and DoD Directive 2311.01.
Changes from the 2012 directive
The 2023 reissuance replaced "manned/unmanned" terminology with "remotely operated/with onboard personnel"; limited applicability to armed platforms, excluding unarmed surveillance systems; clarified that weapon systems with autonomous modes are covered even if they also offer semi-autonomous modes; integrated the 2020 DoD AI Ethical Principles with new AI-specific design requirements; established the Autonomous Weapon Systems Working Group and clarified USD(P) as the submission channel, adding a flowchart for review determination; added a re-approval requirement for substantial modifications; and added post-fielding monitoring requirements. It also softened some modal language, changing "shall" to "will" and "ensure" to "sufficient confidence," a change that has been contested.
Reactions
Human Rights Watch and Harvard Law IHRC (February 2023)
In a review document issued three weeks after the directive, Human Rights Watch and the Harvard Law School International Human Rights Clinic (IHRC) argued that its standards are too weak. They identified a definitional weakness, in that "appropriate levels of human judgment" is undefined and the 2023 text drops "control"; a lowered threshold, citing the shift from the mandatory "shall" to the permissive "will" and from "ensure" to "sufficient confidence"; and a narrowing of civilian protection, in that "unintended engagement" now focuses on force against unintended targets and excludes indirect civilian harm via infrastructure damage. They also pointed to the track record, noting that no senior review has ever been triggered and characterizing the bar as functionally vacuous. The remedy they sought was an international, legally binding treaty prohibiting autonomous weapons without meaningful human control, a ban on anti-personnel autonomous weapon systems, and a government-wide rather than DoD-only policy.
ICRC position on autonomous weapon systems
The International Committee of the Red Cross, custodian of international humanitarian law under the Geneva Conventions, set out its position on autonomous weapon systems in 2021 and reinforced it in 2024. The ICRC calls for a new legally binding international instrument that prohibits unpredictable autonomous weapon systems (those whose effects cannot be sufficiently understood, predicted, and explained, including, per Spoljaric 2024, those powered by "opaque machine learning algorithms"); prohibits autonomous weapon systems designed or used to apply force against persons, on combined international-humanitarian-law and ethical grounds; and strictly regulates all other such systems through limits on the type of target (military objectives by nature), the duration, geographical scope, and scale of use, the situations of use (for example, civilian-absent environments), and human–machine interaction requirements ensuring effective supervision, timely intervention, and deactivation.
The ICRC reinforced its position in a March 2024 submission to the UN Secretary-General pursuant to UNGA Resolution 78/241 (December 2023, adopted with 152 States in favour) and in President Mirjana Spoljaric's statement of 3 May 2024 at the Vienna Conference on Autonomous Weapon Systems 2024, calling for a binding instrument to be negotiated by 2026.
Directive 3000.09's "appropriate levels of human judgment" is narrower than the ICRC's "meaningful human control," operates at the national-policy level rather than as binding international law, and prohibits no category of autonomous weapon. The two approaches diverge structurally: treaty prohibition versus national regulation, meaningful human control versus appropriate human judgment, and categorical prohibitions versus permission with review.
Relation to other governance approaches
Directive 3000.09 functions as the U.S. governance instrument for lethal autonomous weapon systems (LAWS) in practice: a national military-policy regime rather than an international treaty. The UN Convention on Certain Conventional Weapons (CCW) has discussed LAWS since 2014 without producing binding regulation, and the directive operates as the U.S. substitute for a non-existent international framework. Among the Autonomous Weapons governance landscape, it is the principal U.S. institutional-policy instance, set against advocacy from Human Rights Watch, the ICRC, and the Campaign to Stop Killer Robots for categorical prohibitions on certain classes of weapon.
Anthropic's contract terms have been stricter than the directive requires. As described in Clawed, Anthropic's classified Department of War (DoW) contract restrictions barred mass surveillance on Americans and lethal autonomous weapons, categorically excluding Claude from LAWS use, whereas the directive permits LAWS subject to senior review. Dean Ball's analysis characterized these as policy constraints imposed through a contract vehicle rather than regulatory requirements, and described the DoW's retaliatory "supply chain risk" designation as disproportionate.
Relationships
- instance-of: Autonomous Weapons — the US institutional-policy instance.
- supersedes: DoD Directive 3000.09 (Nov 21, 2012, incorporating Change 1 May 8, 2017) — formally reissued and cancelled by this version.
- related: Clawed — Anthropic-DoW conflict plays out against (and stricter than) this directive.
- related: America's AI Action Plan — DoD AI acquisition under the 2025 Action Plan sits downstream of this directive for weapon systems.
- contradicts: ICRC Position on Autonomous Weapon Systems (source summary) — national-regulatory vs. international-prohibition approaches; "appropriate human judgment" vs. "meaningful human control."
- related: International Committee of the Red Cross (ICRC) — the issuing entity of the contradicting international position.
- contradicts: Human Rights Watch / Harvard Law IHRC 2023 Review (to be ingested) — sees directive's standards as too weak.
- depends-on: DoD's 2020 AI Ethical Principles (Responsible, Equitable, Traceable, Reliable, Governable).
- depends-on: DoD Directive 5025.01 (10-year reissuance cycle).
- related: AI and Authoritarianism — "appropriate human judgment" is the US operational response to autonomous-force concerns.