The Smart Marty Sky Dispatch
Sweden's Saab Lifts Lid on AI Fighter Agent Development
Developed in-house, the AI agents have been trained to perform a variety of air combat tasks
Sweden’s Saab has publicly disclosed progress on an in-house program to develop AI agents intended for air combat tasks, according to Aviation International News. The company’s announcement centers on the creation of artificial intelligence agents that can perform a variety of air combat duties, with Saab emphasizing that the work is an internal effort rather than a collaboration tied to external partners for the core AI development.
AeroTime adds detail to Saab’s statement, reporting that the AI agents have been flight-tested on a Gripen E fighter, with at least one agent taking control of the aircraft during beyond-visual-range (BVR) combat scenarios. The outlet notes that Saab previously used an external partner for AI development in earlier tests, but the latest tests involve AI agents developed entirely in-house by Saab, along with the training environments and tools Saab employed to build them.
AeroTime recounts Saab’s chronology, mentioning that prior demonstrations in May and June 2025 used Centaur, an AI agent created by German defense software company Helsing, as part of Saab’s Project Beyond. The current tests, according to AeroTime, rely on Saab-developed agents and the accompanying in-house training infrastructure, under Saab’s Advanced Programs business unit via the AI Fighter Pilot Academy.
According to AeroTime, Saab describes most of the agent training as occurring in simulators, where agents work through diverse air combat scenarios and accumulate millions of simulated flight hours. The ongoing program then moves from simulation to the air, with Saab stating that one of its agents successfully controlled a Gripen E during BVR combat scenarios, although Saab has provided limited specifics about flight dates, the number of sorties, or the opposing aircraft involved.
Saab highlights a software-centric approach to integration, noting that the Gripen E’s software architecture allows new software to be added without major hardware changes. This capability enables direct testing of AI within an in-service fighter rather than requiring a separate experimental aircraft, according to AeroTime’s coverage. Saab frames this approach as enabling iterative AI experimentation within an operational platform.
The AI fighter pilot program is described by Saab as part of broader applications, ranging from assisting cockpit crews to potentially controlling unmanned aircraft operating alongside manned fighters. Saab’s description underscores a spectrum of potential uses for AI agents in combat scenarios, without specifying deployment timelines or operational commitments.
AeroTime attributes the reporting to Saab’s public communications, noting that the agents were developed by Saab’s Advanced Programs unit and the AI Fighter Pilot Academy, with significant training conducted in simulation before transitioning to air testing. The reporting emphasizes that Saab’s in-house development aims to accelerate AI capability while leveraging the Gripen E’s adaptable software framework.
Saab also references its involvement in the broader national AI ecosystem, mentioning Sferical AI, a Swedish sovereign AI initiative in which Saab is a founding partner, as a contributor to the computing resources and AI infrastructure that support agent development and training. Saab’s leadership statement is attributed by AeroTime to Peter Nilsson, head of Advanced Programs in Saab’s Aeronautics division, who asserts that Saab has both built its own AI agents and conducted flight operations with them.
The combined reporting from Aviation International News and AeroTime presents Saab’s lidar on the AI fighter agent project as an ongoing internal program with flight-testing milestones reached, but without providing granular flight data or a detailed disclosure of test dates, sorties, or opponent configurations. The outlets align on the central premise: Saab has progressed from simulator-based training to real-world air testing of in-house AI agents on a Gripen E, with at least one instance of AI control during BVR scenarios.