AI in Military Operations: Autonomous Systems, ISR, and the Future of Military Decision-Making

The gap between what AI can do in a military context and what European militaries are currently deploying has never been more visible — or more consequential. The war in Ukraine has compressed the AI innovation cycle to two-to-four weeks. European headquarters are still measured in years. This guide covers the full operational picture: what AI is doing on the battlefield today, what European militaries are doing about it, and what the technology's development trajectory means for doctrine, law, and strategy.

For European defence professionals and policymakers, understanding this landscape is not a technical curiosity — it is a strategic necessity. The decisions being made now about how to integrate AI into military operations, what doctrine to build around it, how to regulate it, and what capability gaps to close will shape European security for a generation.

The Four Operational Domains Where AI Is Already Active

ISR — Intelligence, Surveillance, Reconnaissance. AI-enabled sensor fusion is transforming the speed and granularity of battlefield intelligence. Computer vision systems process satellite and aerial imagery at volumes that human analysts cannot approach. Pattern-of-life analysis — tracking movement patterns over time to identify military activity, supply routes, and command node locations — is a domain where AI's ability to process large datasets produces outputs that change operational planning. Target identification systems are increasingly capable of distinguishing military objects, vehicles, and positions from civilian ones, though the reliability of these systems in complex environments remains a defining limitation.

Command and control. Decision-support systems are changing how commanders receive, filter, and act on battlefield information. AI systems can ingest feeds from multiple sensor types, cross-reference against known enemy order of battle, generate threat assessments and courses of action, and present synthesised options to human commanders in a fraction of the time the traditional intelligence cycle requires. The human remains the decision-maker — but what the human decides with, and how quickly, is being transformed.

Logistics and maintenance. Predictive maintenance — AI systems that analyse sensor data from military vehicles and equipment to predict failure before it occurs — is one of the domains where military AI is both most mature and most unambiguously beneficial. AI-enabled supply chain optimisation is reducing the logistics burden of complex operations. These applications attract less strategic attention than autonomous weapons, but their operational impact is significant and their adoption is accelerating.

Cyber. AI-enabled offensive and defensive cyber operations represent a domain where the pace of AI development is directly shaping the threat environment. Adversaries are using AI to accelerate intrusion, enhance social engineering at scale, and automate attack sequences. Defensive AI must operate at the same pace — which is why autonomous response capability, not just detection, is now a requirement for serious defence cyber operations.

Autonomous Systems — The Taxonomy That Matters

The public debate about autonomous weapons often conflates several distinct categories of system that have very different operational, legal, and ethical profiles. Getting the taxonomy right is essential for productive policy and procurement decisions.

Loitering munitions — systems like the Switchblade or Shahed-136 — are launched, navigate to a target area, and either loiter waiting for a target or navigate autonomously to a pre-designated location. Most current systems retain a human in the decision loop for final engagement, but the degree of automation varies significantly.

Fully autonomous weapons systems (LAWS) — systems that select and engage targets without human intervention — do not currently exist at operational scale, but the technical components required to build them do. The policy debate about whether to prohibit them, regulate them, or allow their development is active at the UN and within European institutions.

Human-on-the-loop systems — systems where a human can override but does not actively approve each action — are deployed today in a range of contexts, including air defence systems. The speed at which modern threats operate (hypersonic missiles, drone swarms) creates operational pressure to expand human-on-the-loop configurations.

Human-in-the-loop systems — where a human actively approves each consequential action — represent NATO's current stated preference for AI-enabled targeting. The practical challenge is that the operational tempo emerging from the Ukrainian battlefield makes purely human-in-the-loop architectures increasingly difficult to sustain at scale.

Why the distinction matters: legally, because international humanitarian law's requirements for distinction, proportionality, and precaution in attack apply differently to each category; operationally, because the appropriate human oversight architecture varies; and politically, because European defence AI policy is built on commitments about human control that must be operationalised in system design.

The Ukraine Benchmark — What Battlefield AI Actually Looks Like

Ukrainian forces have created the most operationally intense AI development environment in European history. The constraints — denied communications, GPS jamming, electronic warfare, high operational tempo, direct adversary feedback within hours — have driven innovation that no laboratory environment could replicate.

FPV (first-person view) drone guidance systems now incorporate AI navigation that allows operation in GPS-denied environments. Computer vision systems on the drone identify and track targets using onboard processing, eliminating the need for a communication link that can be jammed. Artillery targeting assistance — AI systems that process sensor data to recommend targeting solutions — has become standard in Ukrainian units.

The 2–4 week innovation cycle that characterises Ukrainian battlefield AI development is the product of a specific operational structure: engineers embedded with front-line units, direct access to the environments where systems will be used, real-time feedback from operators, and the freedom to iterate without a procurement bureaucracy in the chain. This is the benchmark against which European military AI development must be honestly assessed — and the structural gap it reveals is the most important capability challenge European institutions face.

ISR at Scale — The Surveillance Architecture Question

Modern ISR generates data at volumes that fundamentally change the nature of the intelligence challenge. A single Sentinel-2 satellite pass generates terabytes of multispectral imagery. A persistent area surveillance system covering a 100km² operational zone — using a combination of fixed sensors, aerial platforms, and satellite passes — generates continuous data streams that no human analyst team can process in tactically relevant timescales.

AI is the enabling layer. Without AI-powered fusion, filtering, and prioritisation, modern ISR data volumes produce not intelligence but noise. The practical requirement is for AI systems that can ingest multiple sensor streams simultaneously, identify objects of interest, flag changes from previous states, cross-reference against known order of battle, and produce analyst-ready outputs — all in timescales that match the operational rhythm of the forces being supported.

European ISR capability has historically been dependent on US intelligence sharing — SIGINT feeds, satellite imagery, and processed intelligence products. The strategic imperative to develop autonomous European ISR capacity is driven not by a desire to duplicate US capability but by the recognition that US intelligence sharing is conditional, selectively available, and potentially unavailable in scenarios where European and US interests diverge.

Decision-Support AI — The Command Level Challenge

The integration of AI into command decision-making is the area of greatest institutional sensitivity — and the area where the risk of getting it wrong is highest. AI systems that support commanders in processing information faster, generating option sets, and assessing threat probabilities are already deployed at brigade and division level in several European militaries.

The challenge is not technical but doctrinal. Military decision-making at the command level involves the exercise of judgment — weighing operational factors, assessing intent, evaluating proportionality, and taking responsibility for consequential decisions under conditions of uncertainty and time pressure. AI systems can improve the quality and speed of the information available for that judgment. They cannot substitute for the judgment itself, and designing systems that are operationally useful without creating inappropriate reliance or accountability confusion requires sustained doctrinal work that most European militaries have not yet completed.

The human-machine teaming question — what the appropriate division of function is between AI systems and human commanders, and how to maintain meaningful human agency as AI decision-support becomes more capable — is the defining institutional challenge of military AI integration. It is neither a technology problem nor a pure policy problem. It requires sustained engagement between technologists, operators, legal advisers, and commanders working from actual operational experience.

Legal and Ethical Constraints — International Humanitarian Law in the AI Age

International humanitarian law — specifically the principles of distinction (between combatants and civilians), proportionality (of expected harm to anticipated military advantage), and precaution (taking all feasible steps to minimise civilian harm) — applies to AI-enabled military operations with full force. The question is not whether IHL applies but how it applies in practice to systems that operate at speeds and scales that change the nature of the decisions involved.

The Campaign to Stop Killer Robots — the coalition of NGOs advocating for a pre-emptive prohibition on lethal autonomous weapons — has pushed the issue onto the UN agenda and generated meaningful political attention in European capitals. Several EU member states have publicly committed to maintaining meaningful human control over lethal force. What "meaningful human control" requires in system design is the technical question that operationalises these commitments.

For European defence AI companies, the IHL question is not primarily a legal compliance issue — it is a market access question. Systems that cannot demonstrate IHL-compatibility will not be procured by European defence forces operating under these commitments. Building IHL analysis into the design process, not the compliance review, is the approach that serious European defence AI companies are taking.

What European Militaries Are Doing

Germany's AI in defence roadmap — the Bundeswehr Digitalisierungsstrategie — focuses on logistics and maintenance AI, decision-support systems at the command level, and ISR data processing. Germany's approach has been cautious and bureaucratically careful, reflecting both the scale of the institutional change required and the specific political sensitivity around autonomous weapons in the German defence context.

France's Ministry of Armed Forces AI strategy — articulated in the 2019 AI and Defence report and updated since — is more operationally ambitious, covering autonomous systems, AI-enabled targeting, and ISR fusion. France's SCAF programme and its broader defence industrial strategy reflect a conviction that maintaining autonomous European AI capability in these domains is a strategic necessity.

Nordic and Baltic approaches — particularly Finland, Estonia, and Norway — tend toward pragmatic adoption of commercial AI capabilities with a specific focus on the threat environment immediately relevant to their geographic position. Estonia's e-defence infrastructure and its relationship with Ukrainian battlefield innovation are particularly notable as a model for rapid AI adoption in a small-state defence context.

The 2027 Summit's Autonomy Track

The autonomy track at the 2027 AI in Defence Summit will address the questions that remain most contested after three years of intense operational experience. What does the Ukrainian benchmark actually imply for European military AI doctrine — and what are the specific institutional barriers to closing the gap? Where does the boundary between human-on-the-loop and human-in-the-loop need to be drawn in European IHL commitments, and what does that mean for system architecture? How does NATO's emerging certification framework interact with national doctrine and procurement decisions on autonomous systems?

These are not questions with clean answers. But they are questions that European militaries, policymakers, and defence AI developers need to be working through together — and the 2027 Summit is designed to be the convening point where that work happens.

See also: