"We also have to have a reality check," Adm. Frank Bradley told attendees at SOF Week in Tampa, Fla., underscoring a cautious approach to introducing artificial intelligence into combat operations.
Adm. Frank Bradley’s caution at SOF Week
Addressing a conference of special operations professionals on Tuesday, Adm. Bradley acknowledged that AI is "changing and impacting and making SOF [special operations forces] more efficient in many ways." He immediately followed that acknowledgment with a warning: of the systems in use on today’s battlefields, "very few, if any of them, are actually using true AI at the edge." Bradley did not rule out the possibility of true edge AI in the future — "We absolutely believe it is" — but he argued the military must be deliberate in how such capabilities are integrated into the "delivery of the battlefield."
Law of Armed Conflict and the limits of automation
Bradley anchored his caution in legal and ethical constraints. He emphasized that "the foundations upon which we go to war, upon which we use violence, are based upon the Law of Armed Conflict." From that premise he drew a clear human-centered boundary: "That is the commander, the person, the human that decides to use lethal violence has to have trust and confidence that that lethal violence will be delivered in the confines of the Law of Armed Conflict, with distinction, proportionately and in the bounds of humanity." He punctuated the point with a simple observation about accountability: "Machines... can't be held accountable."
Technology examples Bradley used: FPV drones, edge AI, and a history of weapon innovation
Bradley placed AI’s potential inside a long sweep of weapons development, listing a progression from "the first stone and sling thrower to the bow and arrow, the machine gun, to the aviation-delivered ordnance, to the Tomahawk land attack cruise missile" and then to "what might come tomorrow with FPV [first person view drones] and edge AI-driven targeting." His comparison suggested both continuity — technologies have repeatedly altered how violence is delivered — and a need for prudence as new tools are introduced.
Consistent with that prudence, Bradley said the current trials and standards used to validate emerging technologies "will remain consistent and critical in the employment of AI in the future." In other words, experimentation should not bypass established validation processes.
Personnel: "PhDs' who can win a bar fight" and "Geeks with Guns"
To exploit technological advances while maintaining control, Bradley said USSOCOM must recruit operators who can handle emerging technology types, including AI. He revived a well-worn characterization of special operators and reframed it for the digital age: his command needs "PhDs' who can win a bar fight." He elaborated with a colloquial turn: "This digital age translates into maybe a more colloquial term: 'Geeks with Guns.' Again plenty of y’all out there, operators who are both lethal but also technically fluent, who can employ cutting-edge tools and the software running it and understand."
The demand is explicit: technical fluency among operators is not optional if AI and similar systems are to be used responsibly and effectively.
What this means for technologists, policymakers, and warfighters
- Technologists and security teams: Bradley’s remarks signal that technologies marketed as "AI at the edge" will face rigorous operational validation before being fielded; developers can expect established trial standards to be applied and should design systems with explainability and compliance in mind.
- Policymakers and regulators: The legal emphasis on the Law of Armed Conflict places policy questions at the center of any expansion of autonomous or semi-autonomous capabilities; Bradley’s framing suggests policy will need to ensure a human decision-maker retains the ability to apply distinction, proportionality and humanity.
- Warfighters and USSOCOM leadership: Recruiting and training priorities will likely continue to emphasize hybrid skill sets — operators who are combat-capable and technically fluent — because "machines... can't be held accountable" and humans must remain "in the loop of that delivery of violence."
Bradley concluded with a reminder that humility must accompany technological ambition: "No algorithm perfectly models any ‘black swan’, and the uncertainties of warfare remain something that no machine can predict." The public record from SOF Week shows a command ready to harness AI’s benefits but determined to preserve human judgment, legal compliance, and rigorous testing as prerequisites for turning new algorithms into battlefield effects.
