The Risk of Decision Compression
Artificial intelligence (AI) is reshaping modern warfare by accelerating intelligence collection analysis, threat identification and military decision-making across land, maritime, air, cyber, and space operations. By processing large volumes of data in real time, AI can substantially shorten the period between detecting a potential threat, deciding how to respond, and acting. This increased speed offers important operational advantages, particularly in complex and contested environments.
However, it also creates a fundamental challenge. AI systems can generate assessments and targeting recommendations faster than human commanders can properly evaluate and verify them. Unless military institutions adapt their decision-making procedures, legal safeguards, and command structures, human oversight risks becoming a mere formality rather than a real check on an AI-generated recommendations. This increases the risk of miscalculation, targeting errors, and unintended escalation.
AI in Recent Conflicts
Recent conflicts have demonstrated the increasingly prominent role played by artificial intelligence in accelerating military decision-making. In the Russia–Ukraine war, both sides have combined AI-assisted analysis, commercial satellite imagery, and unmanned aerial systems to process information and identify potential targets more rapidly.
In the Gulf, the 2026 war involving the United States, Israel and Iran has brought the issue into sharper focus. US officials said the Pentagon used the Maven Smart System to identify priority targets and help select weapons, in a campaign that struck more than 13,000 targets in its first 38 days.
The Gulf’s experience of defensive decision-making is starker still. By 10 May, the UAE Ministry of Defence reported that its air defences had intercepted 551 ballistic missiles, 29 cruise missiles and 2,265 drones fired from Iran. At that volume, human approval of every engagement would be operationally impractical. This does not necessarily indicate a failure of oversight. Indeed, for defensive systems, human control may depend more heavily on system design and rules of engagement. In offensive targeting, however, the central question is whether the person authorising an action can understand and trust the AI system in question.
Information Integrity and Black-Box Systems
Human judgement is only as good as the information it receives, and in contested environments that information is itself a target. Military operations are susceptible to data poisoning as well as interference with navigation signals, deceptive communications, cyberattacks against critical infrastructure, and dependence on commercial satellite services such as Starlink. Meanwhile, AI-generated images and videos, automated disinformation, and coordinated influence campaigns can further distort the information environment . Such manipulation can compromise the data on which AI systems rely, producing unreliable outputs.
A separate challenge is the black-box problem. The complexity of modern deep-learning systems can make their outputs difficult to explain, even when their inputs are known. In high-risk military operations, understanding how an assessment was generated is important because commanders need to evaluate the evidence and assumptions behind a recommendation.
Explainable AI systems designed to make system outputs more understandable, can support this process. However, explainability alone cannot guarantee accuracy or prevent deliberate data manipulation. Generative AI may produce convincing but inaccurate conclusions, sometimes referred to as “hallucinations”, while image- or signal-recognition systems may generate false positives. Effective safeguards therefore require explainability alongside rigorous testing, independent assurance and red-teaming to identify vulnerabilities. Systems must also communicate uncertainty in ways commanders can use under time pressure in order to mitigate the risk of friendly fire, civilian harm, and unintended escalation.
Psychological pressure
These technological vulnerabilities are intensified by the psychological pressures placed on human decision-makers. When AI systems generate large volumes of information and recommendations within a short period, commanders may experience cognitive overload. Under these conditions, they may develop ‘automation bias,’ meaning an excessive tendency to trust and accept a system’s recommendation without sufficient independent scrutiny.
The problems of cognitive overload and automation bias are related. Commanders may be unable to trust a recommendation because they cannot understand its basis, yet may also over-trust the system under pressure. The underlying challenge is knowing when to rely on AI and when to question it. Commanders may retain formal authority over the use of lethal force while lacking the time, contextual understanding or technical knowledge to exercise that authority meaningfully.
Preserving effective human control will require more than keeping a person formally involved in the decision. It will also require sufficient time, reliable information, clear explanations, and the institutional authority to question or reject an AI-generated recommendation.
Institutional Transformation Beyond Technical Procurement
Responding effectively to the increasing speed of AI-supported military decision-making requires broader institutional transformation. Artificial intelligence should not be treated simply as another item of hardware or software to be purchased and added to existing systems. Its responsible integration may require changes to organizational structures, procurement processes, command arrangements, and military doctrine. As future conflicts are likely to involve efforts to manipulate or disrupt information as well as attacks against physical assets, defence institutions need to also protect the systems through which data and software are developed, stored, updated, and shared.
Technological adoption must be accompanied by sustained investment in personnel and professional expertise. Contemporary armed forces require commanders who understand the capabilities and limitations of AI, alongside software specialists, operators of uncrewed systems, and multidisciplinary teams capable of independently evaluating AI-generated assessments during military operations. This does not mean that every commander must become a technical expert. Rather, decision-makers must possess sufficient knowledge to question automated recommendations, recognize uncertainty, and understand when human judgement should take precedence.
Maintaining operational readiness will therefore require continuous professional education, adaptable leadership practices, and stronger cooperation among defence institutions, technology companies, and universities. Such collaboration can support innovation while ensuring that new systems are evaluated against operational, legal, and ethical requirements before they are used in high-risk environments.
Strategic Outlook
Managing the risks created by increasingly rapid AI-supported decision-making, often described as ‘decision compression,’ will require a carefully phased programme of institutional transformation. Defence organizations seek to benefit from the speed and analytical capacity of AI without weakening human judgement, legal accountability, or strategic stability.
First, defence leaders should prioritize investment in explainable AI, which enables users to understand how a system reached a particular conclusion. This should be supported by AI assurance, which involves the systematic testing and evaluation of AI systems to ensure their continued reliability, security, and suitability for their intended purposes. Clear governance arrangements need to define who has the authority to approve AI-supported actions, how sensitive data will be protected, and when human review is mandatory. These safeguards are particularly important when AI-generated recommendations could contribute to the use of lethal force.
Second, military institutions need to modernize how information is collected, assessed, and shared. This will require improving intelligence processes and developing secure command-and-control systems capable of supporting timely coordination across military services and operational areas. However, technological speed should not become the sole measure of effectiveness. Systems must also provide commanders with sufficient context, indicate uncertainty, and allow time for informed human judgement.
Finally, successful adaptation will depend on institutional change rather than technology alone. Armed forces will need flexible organizational structures, appropriately trained personnel, and closer cooperation with government agencies, allied militaries, universities, and the private sector.
The strategic objective should not be to remove humans from military decision-making, but to ensure that human judgement remains informed, responsible, and effective within an increasingly automated environment. Defence institutions that achieve this balance will be better positioned to respond to modern security threats without compromising legal and ethical responsibility.
