Official policy
Strong evidence for stated governance, program purpose, and public requirements—not necessarily field performance.
Public evidence through July 31, 2026
Who is using AI-enabled kill-chain functions—and how
The public record shows widespread adoption of AI-assisted sensing, fusion, decision support, autonomous mobility, terminal recognition, and defensive automation. It rarely provides a complete account of software, rules of engagement, human intervention, or deployed configuration.
Official facts separatedManufacturer claims labeledClassified details remain unknown
Official program description
Strong evidence that a program or capability exists, with limits where implementation details are classified.
Manufacturer statement
Evidence of what a supplier publicly claims; not independent confirmation of operator doctrine or combat results.
Independent analysis
Contextual interpretation based on public material, with source quality and uncertainty varying by topic.
North America
United States
Public U.S. evidence is strongest for AI-assisted intelligence, sensor fusion, target cueing, command-and-control, autonomy research, and bounded autonomous functions—not a blanket transfer of lethal authority to AI.
Publicly evidenced uses
- Maven Smart System: fuses sensor data for object detection, tracking, and decision support.
- CJADC2 and related programs: connect distributed sensors, decision nodes, and effectors across domains.
- Replicator: accelerates fielding of attritable autonomous systems, with authority and mission modes varying by system.
- Missiles, air defense, and collaborative aircraft use bounded navigation, sensing, classification, or mission-autonomy functions.
Human control and governance
DoD Directive 3000.09 requires appropriate levels of human judgment, realistic testing against adaptive adversaries, safety and cybersecurity, understandable interfaces, and legal compliance.
Unknowns and limits
Classified rules, exact model architectures, operational thresholds, and weapon-release modes for many systems are not public.
Evidence class
Official policy and program evidence
Middle East
Israel
Israeli manufacturers publicly describe systems spanning automatic target recognition, human-in-the-loop precision weapons, and bounded autonomous anti-radiation loitering munitions.
Publicly evidenced uses
- SPICE family: scene matching, automatic target acquisition/recognition, autonomous functions, and a data link supporting human-in-the-loop updates and mission abort.
- HARPY: manufacturer-described autonomous search and attack against emitting radar targets within a defined mission profile.
- HAROP: electro-optical loitering system publicly differentiated from HARPY through human-in-the-loop control material.
- AI-assisted target-generation systems have been widely reported, but claims about workflows, error rates, and human review are disputed and incompletely public.
Human control and governance
Control arrangements differ by system and mode; “autonomous” marketing language must be parsed at the function–context level.
Unknowns and limits
Operational software, thresholds, deployment rules, review procedures, and many conflict-specific claims are not fully public or independently verifiable.
Evidence class
Manufacturer descriptions plus public reporting
Europe
Norway
Norway’s clearest public example is the Naval Strike Missile family, where autonomous target recognition supports terminal discrimination after a human-defined mission.
Publicly evidenced uses
- Imaging-infrared sensing and autonomous target recognition for terminal target discrimination.
- Networked coastal- and ship-based command systems connect surveillance, fire control, and missile launchers.
- The same missile family is selected by multiple allied operators, making Norway a significant exporter of bounded post-launch autonomy.
Human control and governance
Public material establishes mission planning and terminal recognition but does not fully disclose intervention and release logic for every operator and configuration.
Unknowns and limits
Exact software, classifier performance, abort behavior, and country-specific rules of engagement are not public.
Evidence class
Manufacturer and public operator evidence
East Asia
People’s Republic of China
PLA writings and external research emphasize “intelligentization,” human–machine integration, distributed sensing, decision advantage, swarms, and kill-web concepts. Public operational details are limited.
Publicly evidenced uses
- Research and doctrine concerning AI-supported command decisions and multi-domain sensing.
- Development of unmanned and swarm systems through military–civil fusion and a broad commercial technology base.
- Interest in algorithmic decision speed and cognitive-domain operations.
Human control and governance
Publicly accessible governance and weapon-release rules are less transparent than those published by the United States or NATO.
Unknowns and limits
The maturity, combat validation, control modes, and fielded scale of particular AI-enabled kill-chain functions are often unknown.
Evidence class
Doctrine-focused external research; limited operational transparency
Europe / Eurasia
Russia
Russia publicly promotes AI-assisted drone recognition, terminal autonomy, electronic-warfare resilience, and reconnaissance–strike integration. The exact degree of autonomous target selection is frequently unclear.
Publicly evidenced uses
- ZALA publicly describes intelligent targeting and recognition in Lancet-family material.
- Drone ecosystems increasingly use onboard processing to operate when communications or navigation are degraded.
- AI is also used in counter-drone sensing, imagery interpretation, and command support.
Human control and governance
Public manufacturer language does not reliably establish who authorizes each engagement, how abort works, or how models behave outside advertised conditions.
Unknowns and limits
Combat performance, error rates, software versions, operator oversight, and autonomy modes are contested or undisclosed.
Evidence class
Manufacturer claims, wartime reporting, and independent analysis
Europe
Ukraine
Ukraine has rapidly adopted AI-assisted detection, navigation, targeting support, and terminal guidance to cope with electronic warfare and the scale of drone operations.
Publicly evidenced uses
- Public reporting describes Saker Scout AI-assisted detection and identification functions.
- Onboard vision and navigation are used to reduce dependence on continuous operator links in jammed environments.
- AI also supports imagery triage, damage assessment, logistics, and counter-drone defense.
Human control and governance
Different systems may be remotely piloted, supervised, or terminally autonomous; “AI-enabled” should not be read as proof of fully autonomous weapon release.
Unknowns and limits
Rapid field modification, classified software, inconsistent naming, and limited independent access make exact control arrangements difficult to verify.
Evidence class
Government-linked announcements, public reporting, and battlefield observation
SourcesUkrainska Pravda 2023-09-04
Europe
Estonia
Milrem Robotics provides a clear example of separating autonomous mobility from lethal authority in the THeMIS uncrewed ground vehicle.
Publicly evidenced uses
- Follow-me, return-home, waypoint navigation, and obstacle avoidance.
- Integration with remote weapon stations and other mission payloads.
Human control and governance
Milrem states that autonomous functions are strictly limited to mobility and that the weapon system is controlled by a human operator.
Unknowns and limits
Configurations and operator doctrine vary by customer; manufacturer statements do not independently verify every deployed use.
Evidence class
Manufacturer statement
SourcesMilrem Robotics 2022-06-14
Europe / North America
Germany / Canada
Rheinmetall Canada’s Mission Master family demonstrates autonomous mobility, networked sensing, and automated cueing while reserving kinetic decisions to humans.
Publicly evidenced uses
- Autonomous navigation, convoy, and follow-me behavior.
- A surveillance vehicle can share a detected location and cue a separate fire-support vehicle.
Human control and governance
Rheinmetall states that targets are never engaged automatically and that a human in the loop is required for all kinetic decisions.
Unknowns and limits
Future configurations, customer-specific integration, and fielding scale vary and must be checked individually.
Evidence class
Manufacturer statement and demonstrations
SourcesRheinmetall 2021-08-10
Transatlantic alliance
NATO (alliance framework)
NATO’s public role is best understood as standard setting, interoperability, assurance, and governance across allied AI applications rather than one monolithic “NATO kill chain.”
Publicly evidenced uses
- Alliance-level AI strategy, data governance, experimentation, and interoperability.
- Principles of responsible use applied across defense and security AI applications.
- Work on certification, testing, assurance, and governability.
Human control and governance
NATO lists lawfulness, responsibility and accountability, explainability and traceability, reliability, governability, and bias mitigation.
Unknowns and limits
National systems, rules, and operational authority remain with allies and vary substantially.
Evidence class
Official alliance policy
SourcesNATO 2024-07-10
Cross-country pattern
The most common public use is upstream of weapon release.
Imagery triage, detection, fusion, tracking, planning, navigation, and resource allocation are more widely documented than unrestricted machine selection of people for lethal engagement.
| Pattern | Where it appears | Why it is attractive | Main assurance problem |
|---|---|---|---|
| AI-assisted ISR and target cueing | United States and allies; widely pursued elsewhere | Scale imagery and sensor analysis; reduce analyst workload | Automation bias, provenance, false classification, and review capacity |
| Networked kill webs | United States, NATO members, China-focused research | Connect any suitable sensor to an available effector | System-of-systems validation and cascading track error |
| Autonomous mobility | Estonia, Germany/Canada, United States, Ukraine, others | Operate under workload, distance, and communications constraints | Separating mobility autonomy from lethal authority and lost-link behavior |
| Terminal recognition and guidance | Israel, Norway, United States, Russia, Ukraine | Operate in clutter, under jamming, or without continuous control links | Target-profile bounds, open-set recognition, abort, and disclosure of control mode |
| Automated local defense | Many states | Reaction time against fast materiel threats | False tracks, fratricide, saturation, and escalation compression |