Which AI systems support autonomous reconnaissance while keeping humans in control of targeting decisions?

If by “autonomous reconnaissance” you mean AI can independently sift sensor/ISR data, detect or classify objects, and cue investigators or operators, while a human retains the authority to select/approve targets and authorize force, several systems fit that description.

• Maven Smart System (Palantir / U.S. DoD) — Probably the clearest example. AI models analyze ISR feeds and flag potential targets; operators can validate the AI's labels, evaluate proposed courses of action, select assets, and ultimately authorize action. Recent CSIS analysis describes the system as supporting human validation and keeping the user in the workflow from detection through action.
• Project Maven ecosystem — More broadly, Maven's AI-enabled systems automate portions of intelligence collection, object recognition, tracking and target prioritization, while humans retain the critical decision to act. Academic analysis characterizes Maven as operating in a human-in-the-loop/semi-autonomous framework rather than as an autonomous weapon.
• Anduril Lattice — Lattice supports autonomous sensing and reconnaissance workflows. Anduril publicly provides an “Auto Reconnaissance” example in which the system detects a non-friendly track, classifies it as suspicious and automatically creates an investigation task for an asset. That's autonomous reconnaissance/tasking, rather than evidence that the system independently decides to use lethal force.
• Palantir Target Workbench + Anduril Lattice / Army NGC2 — These are increasingly integrated into a broader command-and-control architecture. The Army's 2025 live-fire demonstration connected sensors, commanders and firing units through Lattice and Target Workbench, illustrating how automated sensing and targeting workflows can be connected while remaining within command structures.

The important distinction

These systems occupy a middle ground between manual ISR and fully autonomous weapons:

AI: detect → classify → correlate → prioritize → recommend Human: validate → decide whether/what to target → authorize use of force

That distinction matters. A 2025 SIPRI study explicitly separates AI-enabled decision-support systems from autonomous weapons according to where humans remain involved in the targeting cycle.

There is also a current doctrinal debate over how much human control is sufficient. A 2026 U.S. Army article proposes a model called “human decide–machine detect–machine deliver–machine assess,” in which commanders retain responsibility for validating targets and approving engagement parameters while machines perform detection and potentially execution within those constraints.

So, if your requirement is specifically “machine-autonomous reconnaissance, but humans make the actual targeting/engagement decision,” Maven Smart System and Lattice are the strongest publicly documented examples. They should not, however, be treated as interchangeable with systems that autonomously select and attack human targets—the latter raises substantially different legal and safety issues.

Several defense AI platforms specialize in autonomous reconnaissance, intelligence processing, and battle management while retaining a strict "human-in-the-loop" or "human-on-the-loop" requirement for applying lethal force , complying with frameworks like the U.S. Department of Defense Directive 3000.09.

• Maven Smart System (developed primarily by Palantir Technologies): Acts as an AI-enabled data fusion and decision-support platform . It ingests massive streams of imagery, sensor data, and reconnaissance feeds to identify, classify, and track potential military targets, routing options directly to human commanders who maintain final authorization over any engagement.
• Anduril Lattice OS (by Anduril Industries ): An AI-powered software platform operating uncrewed aerial and ground surveillance systems . Lattice autonomously manages sensor networks, maps hostile or contested environments, and tracks moving objects, but requires human operators to make explicit targeting or response decisions.
• Project Maven-integrated Foundation Models : Defense contracts awarded to major tech providers (including OpenAI, Anthropic , and Google) focus on integrating secure large-scale reasoning and computer vision models for administrative logistics, threat detection, and reconnaissance synthesis, explicitly constrained from autonomous weapon release.
• Tzayad (Israel's Digital Army Program) : An advanced command-and-control battle management system that automates the collection and processing of battlefield reconnaissance data to suggest courses of action, leaving actual strike validation strictly to human military personnel.

If you'd like, I can dive deeper into:

• The specific legal and ethical frameworks governing autonomous military AI (such as DoD Directive 3000.09)
• The technical difference between human-in-the-loop versus human-on-the-loop systems

Let me know how you'd like to proceed.

If by “autonomous reconnaissance” you mean AI can collect/fuse ISR data, detect objects or anomalies, track them, and generate candidate targets, while a human retains authority over whether a target is actually selected/engaged, several systems fit that model.

• Project Maven / Maven Smart System (MSS) — probably the clearest example. It processes satellite, drone, radar, and other ISR feeds; computer-vision models flag and label potential targets; human operators can review/validate those detections and make targeting decisions. The U.S. Air Force explicitly describes its Maven experiments as using AI to support rather than replace the human decision-maker.
• TITAN (Tactical Intelligence Targeting Access Node) — an Army deep-sensing/ISR system associated with Palantir and Anduril. It fuses space, aerial, terrestrial and other sensor data and automates target recognition/geolocation to provide actionable targeting information to commanders and operators.
• NATO's AI-enabled surveillance/“kill web” architecture — current NATO efforts envision autonomous drones and sensors performing surveillance and threat detection while human operators retain control of lethal actions.
• U.S. Air Force Maven/Maverick human-machine teaming experiments — these use AI to recommend courses of action and accelerate the F2T2EA targeting process, while operators can compare AI recommendations with their own decisions.
• DARPA/USAF autonomous-aircraft programs such as VENOM/AIR are somewhat different: AI can control aircraft and eventually coordinate autonomous aircraft, but the emphasis is on human pilots commanding and orchestrating the autonomous systems rather than independently authorizing lethal targeting.

A useful distinction is human-in-the-loop vs. human-on-the-loop. The former generally requires a person to approve the relevant action; the latter can allow the machine to execute within predefined constraints while a person supervises. Current U.S. military discussions increasingly call the broader concept “guided autonomy”: AI handles sensing, analysis, coordination, and other machine-speed tasks, while humans retain responsibility for consequential decisions, particularly use of force.

So, if your criterion is specifically “autonomous reconnaissance + human-controlled targeting,” Maven/MSS is the strongest publicly documented example, with TITAN and NATO's emerging sensor architectures fitting related versions of the same pattern.

One caveat: public descriptions don't establish that humans exercise identical control in every real-world deployment. The degree of automation and authorization can vary by mission, rules of engagement, and system configuration. This distinction is increasingly important because autonomous targeting systems are now the subject of active international debate.

A growing class of defense and dual-use AI platforms are engineered specifically for autonomous intelligence, surveillance, and reconnaissance (ISR) while structurally maintaining a human-in-the-loop or human-on-the-loop requirement for kinetic or targeting decisions.

These platforms use machine vision and sensor fusion to process the massive data firehose from drones, satellites, and ground sensors, leaving the ultimate authorization of force to human operators.

Key AI Systems Supporting Autonomous Reconnaissance with Human-Controlled Targeting

• Maven Smart System (Palantir / Department of Defense) : Originally born under Project Maven, this AI-enabled capability acts as a massive decision-support engine. It ingests live video feeds, satellite imagery, and signals intelligence to autonomously detect, classify, and flag objects of interest (such as military vehicles or hostile maritime assets). However, the system is explicitly designed to present these options to human analysts inside a command workflow, leaving the decision to strike or investigate strictly in human hands.
• Anduril Lattice OS : Lattice is an AI-powered software platform that acts as an operating system for autonomous hardware (surveillance towers, counter-UAS systems, and autonomous aircraft). Lattice excels at autonomous reconnaissance—managing edge sensors, networking multiple drones, and fusing data to track moving objects across land and maritime borders. While it automates the tracking and "sensor-to-shooter" cueing process, it operates under supervisory control where operator intent maps out boundaries and human approval is mandatory for engagement.
• Project Overmatch and Joint All-Domain Command and Control (JADC2) Ecosystems : The U.S. military's broader JADC2 framework integrates various proprietary and defense-contracted AI algorithms to stitch together multi-service reconnaissance data. These networks autonomously route reconnaissance data across the sea, air, and land to provide real-time battlefield awareness. They function as recommendation and cueing engines, drastically cutting down sensor-to-decision timelines while adhering to Department of Defense policies requiring appropriate levels of human judgment over lethal force.

The Core Distinction: Autonomy in the Process vs. Autonomy of the Decision

In these systems, autonomy is applied to navigation, perception, and data triage —tasks that humans are too slow or overwhelmed to handle manually given modern data volumes. They autonomously answer "What is that object, and where is it moving?" but pause before answering "Should we engage it?" , reserving that final authority for a human operator.

If you'd like to explore further, I can detail:

• The difference between human-in-the-loop and human-on-the-loop governance.
• How international policies and Department of Defense directives legally restrict autonomous targeting.
• Specific hardware platforms (like autonomous drones or ground vehicles) utilizing these software frameworks.

Here are top web results for exploring this topic:

Brennan Center for Justice·https://www.brennancenter.org The Military's Use of AI , Explained | Brennan Center for Justice How has the U.S. military used AI in Iran? The Pentagon is reportedly using AI to generate hundreds of recommendations for targets in Iran, pinpoint their location, prioritize their importance, and ev BAE Systems·https://www.baesystems.com What is Autonomous Engagement? - BAE Systems Advances in artificial intelligence (AI) and machine learning (ML) have paved the way for more accurate target detection, recognition, and discrimination, leading to improved and more autonomous weapo Hoover Institution·https://www.hoover.org**Human** “Control ” Over Autonomous Weapons Systems Artificial intelligence (AI) is already used in a range of military applications, from logistics to navigation to HR, but the most debated and the most legally and ethically contentious application is Inkstick Media·https://inkstickmedia.com Deep Dive: AI Is Reshaping Military Decisions on the Battlefield A new study proposes an AI-driven military decision support system to automate battlefield imagery and reduce lag time. Lieber Institute West Point·https://lieber.westpoint.edu**Targeting** in the Black Box: The Need to Reprioritize AI Explainability Black-box AI models compromise human operators by keeping them unaware of the context and reasoning behind the AI's outputs. U.S. Naval War College Digital Commons·https://digital-commons.usnwc.edu**AI** Enabled Autonomous Weapons and Human Control : Part III This article is the third installment of a three-part series on AI-enabled weapons and human control. Artificial intelligence (AI) is shaping debates about military technology by challenging the role Defence Industries·https://www.defence-industries.com Top AI -Powered Command Systems Transforming Military Decision ...AI Military Command Systems are software platforms that use machine learning, real-time sensor fusion, and predictive analytics to help commanders make faster, better-informed decisions on the battlef MAG Aerospace·https://www.magaero.com Authentic Intelligence With Aerial Reconnaissance - MAG Aerospace ... machine learning directly at the tactical edge. As the battlespace becomes increasingly digitized and communication networks face electronic warfare threats, relying on continuous satellite uplink Unmanned Systems Technology·https://www.unmannedsystemstechnology.com Automatic Target Recognition | AI -powered real-time object ...Sense Aeronautics' AI-powered Automatic Target Recognition (ATR) solution automatically detects and classifies targets of interest in EO and IR video streams, delivering results that are faster, more Reddit·https://www.reddit.com**Autonomous** weapons guided by artificial intelligence are already in ...The emergence of AI on the battlefield has spurred debate among researchers, legal experts and ethicists. Some argue that AI-assisted weapons could be more accurate than human-guided ones, potentially

Yes. If by “autonomous reconnaissance” you mean AI can independently collect/analyze ISR data, detect or classify objects, track them, and recommend/prioritize targets—but a human retains the decision to engage, several systems fit.

The clearest examples

DARPA URSA (Urban Reconnaissance through Supervised Autonomy) is almost exactly the architecture you describe. Its purpose is to have autonomous systems perform reconnaissance and distinguish hostile from innocent people in complicated urban environments, while ground forces supervise them. DARPA explicitly specifies that humans make the final escalation decision, with the autonomous system providing additional intelligence beforehand.

Palantir Maven is another important example, although it is better described as an AI-enabled intelligence/targeting system than an autonomous reconnaissance vehicle. Maven processes enormous volumes of sensor data to identify potential targets and support targeting workflows. Recent reporting says humans retain control over lethal decisions.

NATO's Next Generation Targeting (NGT) is perhaps the clearest current institutional example of the human-controlled targeting model. NATO says NGT will automate information synthesis, filter battlefield “noise,” and help track and validate targets, while its human-control interface preserves commanders' authority. NATO explicitly states that technology is intended to support commanders, not replace them.

Anduril Lattice goes further toward autonomous multi-platform operations. In an Army demonstration, a single operator used Lattice to command multiple uncrewed aircraft that autonomously located and identified a SAM site. Its current mission-autonomy architecture is designed to let human operators interact with teams of autonomous systems.

Shield AI Hivemind is somewhat different: its core purpose is mission autonomy. It can independently fly aircraft, adapt to changing conditions and execute coordinated behaviors. The Air Force is now using it in its Collaborative Combat Aircraft program. That makes it a good example of autonomous reconnaissance/mission execution, but not necessarily the cleanest example of “AI recommends a target while a human alone decides whether to engage.”

The broader pattern

The architecture you're describing is generally “human-on-the-loop” or supervised autonomy, rather than fully autonomous weapons:

Sensors → autonomous collection/fusion → AI detection/classification → target recommendation/prioritization → human validation/decision → engagement

That distinction matters. NATO's AI policy explicitly calls for appropriate human-machine interaction, human responsibility/accountability, and the ability to disengage systems exhibiting unintended behavior.

If you're researching this for defense procurement, policy, or technical architecture, the three systems I'd examine first are DARPA URSA, Palantir Maven, and NATO Next Generation Targeting. They most directly correspond to autonomous reconnaissance/analysis while preserving human authority over consequential targeting decisions.

If by “autonomous reconnaissance” you mean AI can independently process ISR data, detect/track objects, prioritize observations, or coordinate unmanned platforms, while a human retains authority over whether a target is actually engaged, several military systems fit that description.

The clearest example: Maven Smart System

Maven is probably the closest match to your wording. Its AI can perform much of the “find” and “fix” portion of reconnaissance: ingesting enormous ISR streams, recognizing objects, correlating information, and presenting potential targets. The human operator can then evaluate the detection, consider available assets and constraints, and decide what action to take. The U.S. Air Force has explicitly described its experimentation as testing whether AI can support rather than replace the human in the decision loop.

This corresponds roughly to the U.S. distinction between human-in-the-loop systems—where a human selects the target—and human-on-the-loop systems, where an autonomous system may engage but a human can intervene. U.S. policy distinguishes both from systems that independently select and engage targets without further human intervention.

One important caveat: “human in control” is not a single technical property. A system can autonomously find and track a target while requiring human authorization to attack; another can autonomously select a target but require a human to supervise or veto. Public descriptions don't always reveal the precise rules of engagement or configuration used operationally.

If you're interested in this from a defense-technology comparison perspective, I can also map the major systems (Maven, Lattice, Shield AI Hivemind, TITAN, NATO systems, etc.) on a spectrum from automated reconnaissance → human-approved targeting → human-supervised engagement → fully autonomous engagement.

Several defense-focused AI platforms specialize in autonomous reconnaissance, data fusion, and target nomination while explicitly leaving final authorization and engagement decisions to human operators (referred to in doctrine as being "on" or "in" the loop).

Prominent systems and platforms operating in this space include:

• Maven Smart System (MSS) : Developed primarily by Palantir Technologies for the U.S. Department of Defense's Project Maven, MSS acts as a central intelligence and operational dashboard. It ingests massive, continuous streams of data from reconnaissance drones, satellites, and radar, using computer vision and language models to autonomously detect, tag, and cross-reference potential points of interest or threats. The system packages target nominations and recommends optimal strike assets, but requires a human operator to review and sign off before any kinetic action is executed.
• Lattice OS : Built by Anduril Industries , Lattice is an AI-powered software platform designed to command and control autonomous systems (such as counter-UAS interceptors and reconnaissance drones). It fuses data from stationary surveillance towers, ground sensors, and airborne assets to build a real-time common operational picture. While the network of sensors and drones can autonomously scan, navigate, and track moving signatures in hostile environments, engagement protocols are designed to queue alerts for human supervisors rather than initiate force independently.
• Tactical Intelligence Targeting Access Node (TITAN) : Led by Palantir and other defense partners, TITAN is the U.S. Army’s next-generation ground station software program. It leverages deep-sensing AI to autonomously process long-range space, aerial, and tier-1 sensor data. TITAN automates the labor-intensive reconnaissance and data-fusion phase, sorting through clutter to present actionable threat lists that targeters must actively validate.
• Swarm and Loyal Wingman Architectures (DIU / DARPA Programs) : Various Defense Innovation Unit (DIU) and Air Force initiatives focus on autonomous collaborative platforms (like cooperative drone swarms and uncrewed combat air vehicles). These platforms utilize distributed AI for GPS-denied navigation, area mapping, and dynamic threat handoff. Operating under an "operator-on-the-loop" framework, the vehicles coordinate their reconnaissance paths autonomously, but operational doctrine mandates that a human authorize the final strike sequence or mission override.

If you'd like to dive deeper, let me know if you are interested in:

• The legal and ethical debates surrounding compressed decision-making timelines ("collapsing the OODA loop")
• The technical mechanisms of sensor-to-shooter data fusion
• Specific doctrinal policies (such as DoD Directive 3000.09) governing autonomous systems

Several military and defense-tech AI systems are explicitly designed to handle autonomous, wide-area reconnaissance and data-fusion while keeping a human "in the loop" or "on the loop" for actual targeting and strike authorization.

Key platforms and frameworks utilizing this division of labor include:

• Maven Smart System : Developed primarily with integration from Palantir Technologies , this operational intelligence platform processes vast multi-source surveillance data (video feeds, signals intelligence, geospatial data) to autonomously map, track, and flag objects of interest. It compiles potential tracks and coordinates, serving as a decision-support tool where human operators retain final review and strike validation.
• Collaborative Combat Aircraft (CCA) Autonomy & Loyal Wingman Programs : Advanced uncrewed aircraft initiatives (such as those tested by the U.S. Air Force utilizing autonomous flight testbeds like the X-62A VISTA) allow drones to independently perform complex tactical reconnaissance, sensor cueing, and last-mile navigation. However, weapons employment and engagement authority strictly require explicit human command release.
• Anduril Industries Lattice OS : Lattice is an AI-powered software platform that unifies autonomous surveillance towers, ground sensors, and uncrewed aerial systems (UAS). It autonomously builds a real-time situational awareness map and prioritizes anomalous behaviors or threats, presenting them via an operator interface where humans decide on the appropriate response or engagement.
• BAE Systems AI Drone Controllers & Natural Language Mission Systems : These systems allow operators to use high-level natural language or voice commands (e.g., searching a specific quadrant for threats). The onboard or edge AI handles the independent pathfinding, obstacle avoidance, and sensor data processing required to execute the scout mission, leaving target classification confirmation and engagement decisions to the human supervisor.

Would you like to explore the distinction between human-in-the-loop versus human-on-the-loop governance models, or are you looking into a specific domain like maritime, aerial, or cyber reconnaissance?

If by “autonomous reconnaissance” you mean AI can independently collect, process, classify, track, and cue ISR information while a human retains the authority to decide whether a target is engaged, several publicly documented systems fit that general model:

The clearest examples

Anduril Lattice is probably the closest match to your description. Its C2 software autonomously turns large quantities of sensor data into a common operating picture and can task autonomous assets, while its publicly documented EDGE23 demonstration explicitly left the lethal authorization to a human.

Maven Smart System is the other major example. Its emphasis is less on making the physical platform autonomous and more on automating the ISR-to-targeting workflow: AI searches huge quantities of imagery/data, flags objects, and presents information and recommendations to human operators.

A useful distinction is human-in-the-loop vs. human-on-the-loop. In the former, the person must make a decision before the system proceeds with a consequential action. In the latter, the AI may act autonomously within predefined bounds while a human supervises and can intervene. Those aren't equivalent safeguards. Recent scholarship also emphasizes that “human control” needs to be evaluated across the entire AI lifecycle, rather than simply asking whether a human technically has a final button to press.

If your interest is specifically in non-lethal ISR/reconnaissance systems rather than military strike/targeting systems, I can also give you a broader list of platforms used for autonomous surveillance, mapping, maritime reconnaissance, and border/security applications.