From the earliest battlefields to today’s contested skies, victory has often gone to the side that senses and responds the quickest. See the fight, make sense of it, decide, and act before the enemy does. Modern warfare doctrine refers to this as the OODA loop: Observe, Orient, Decide, Act.
The challenge has always been uncertainty – gaps in intelligence, misleading signals, or incomplete situational awareness. Yet, history shows that the ability to move decisively, even in the face of uncertainty, often tilts the outcome in one’s favor. Technology has long played a central role in reducing that uncertainty, speeding commanders through the OODA loop by improving their ability to sense and orient faster.
Over time, enhanced means for sensing and deterministic algorithms have enabled dramatic improvements in the observe and orient portions of the OODA loop. But as both technology and tactics evolve, so does complexity. The modern battlespace now moves at a velocity and scale that can overwhelm traditional human decision cycles. What was once a human advantage or rapid situational understanding can now be outpaced by data overload and adversaries who themselves employ automation and AI.
AI and the acceleration of decision-making
Enter artificial intelligence. AI offers the potential to enhance nearly every aspect of the OODA loop, operating at a scale and speed impossible for human cognition alone. It can parse sensor data, correlate patterns across domains, and generate multiple courses of action all in fractions of a second.
However, AI solutions and their probabilistic basis also leave room to question the level of confidence one can have in the identified “solution,” and then the risk associated with acting on an AI model’s decision. The central challenge becomes how to effectively leverage AI to outpace an adversary’s OODA loop while preserving the ethical and legal imperatives of warfare – ensuring all action remains proportionate and discriminate – while operating “at speed.”
Using human judgment where it counts most
Even as technology advances, some decisions and resultant actions have such immense consequences that human judgment must remain supreme. Nuclear command and control is a crystal clear example: no algorithm, however fast or well-trained, could be trusted with a decision that might end civilization as we know it. Not surprisingly, U.S. policy dictates that the decision to use nuclear force must be made by just one person, the President of the United States.
That said, AI can still be a powerful and increasingly necessary partner, speeding one’s ability to more rapidly execute the OODA loop. In highly fluid situations—say, a commander responding to a complex scenario of adversary actions—AI can sift through sensor data, correlate events, and generate a common operating picture that both summarizes and highlights key elements of the situation. It can then offer multiple courses of action, complete with confidence levels and reasoning. Yet even when AI operates at machine speed, the final decision must remain human.
Engineering trust into autonomy
The probabilistic nature of AI does introduce a new type of operational uncertainty. Models can behave in ways their developers never anticipated, particularly when confronted with conditions beyond their training data. This creates a paradox: AI is invaluable for managing complexity, yet it cannot be guaranteed to respond “correctly” every time. The solution lies in engineering trustworthy autonomy – embedding guardrails that ensure AI systems act predictably and are constrained within acceptable limits. These guardrails must be deliberate, layered, and mission-specific, spanning three essential dimensions: technical, operational, and policy. Explainability, confidence scoring, and real-time anomaly detection are also critical to maintaining operator trust in AI recommendations.
- Technical Guardrails
At the system level, technical guardrails define the boundaries within which AI can act. For example, an aircraft’s AI flight control system may propose evasive maneuvers but remain constrained within the aircraft’s known performance envelope. These boundaries prevent AI from taking actions that could directly endanger the platform or violate mission parameters.
- Operational Guardrails
Even the best-engineered systems must function within disciplined command frameworks. Operational guardrails define when and how AI can act autonomously. A proven example is the Navy’s Close-In Weapon System (CIWS), which can be switched into automatic mode only under clearly defined high threat conditions—when human reaction time would be insufficient to defend the ship. Outside of those narrow circumstances, human oversight is mandatory. Similar operational doctrines should govern the use of AI in other high-risk missions.
- Policy and Ethical Guardrails
Technology and operations cannot exist without policy. The rules of engagement, data governance standards, and procurement guidelines must evolve to explicitly define where AI can act on its own and where human authority must intervene. Sensitive military data must be kept out of public or commercial AI models, and ethical review must be built into the development and deployment of all defense-related AI systems.
Artificial intelligence can be a powerful force multiplier, compressing the OODA loop and empowering commanders with unprecedented speed and insight. But speed must never come at the expense of control.
On the risk-time spectrum, the higher the consequence of error, the greater the requirement for human oversight. Conversely, as time pressure increases, limited autonomy may be warranted, but only within carefully engineered guardrails that ensure decisions remain proportionate, explainable, and accountable. By embedding human, technical, and policy guardrails into every layer of our systems, we can ensure AI serves as a trusted ally rather than an ungoverned risk.