“ability to develop, train, and fly its own Al for air combat,” Saab wrote in a company statement — a short sentence that captures the company’s immediate claim and the technological ambition behind its announcement on November 3, 2026.
Saab’s AI Fighter Pilot Academy: infrastructure, training, and evaluation
Saab says its newly revealed in‑house agents were developed inside an initiative the company calls the AI Fighter Pilot Academy, a program the firm describes as designed to “create, train and evaluate” autonomous agents and the infrastructure those agents require. Peter Nilsson, head of Advanced Programs at Saab’s aeronautics unit, framed the Academy as more than a laboratory: “[T]hat is the true purpose of the AI Fighter Pilot Academy,” he told Breaking Defense and other media ahead of Saab’s announcement.
Nilsson stressed the longer-term challenge: “The harder question is how we turn this success [of flying with the agents] into a dependable capability that can learn, improve, and be relevant over time.” He described the strategic value of shortening the loop among experience, learning, and fielded capability, and highlighted the “enormous value to have a software-driven fighter like Gripen who can cope with this [computer processing] speed.”
Gripen E testing and the focus on Beyond Visual Range (BVR) operations
Saab has run the AI agents primarily on a Gripen E multirole fighter jet. According to the company, the agents use “various Al techniques to master tactical decision-making in complex operational environments” and can operate either alongside a pilot in the cockpit or autonomously from an uncrewed air vehicle.
The firm said testing to date has concentrated on Beyond Visual Range (BVR) combat conditions. Saab’s latest campaign follows a series of flight tests in 2025 when Helsing’s Centaur AI tool was used to take control of the fighter jet and autonomously complete a number of BVR maneuvers — a precedent Saab references directly when describing its current program.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleSferical Al and the computational backbone: DGX SuperPOD and scale
To scale compute capacity for the agents, Saab has become a partial founder in a Swedish government‑backed initiative called Sferical Al. Nilsson explained the strategic rationale in blunt commercial terms: without a domestic option, “otherwise you have to go to Google or Amazon … and pay some money” meaning, “you maybe don’t have full control of your data.” Retaining such capability in Sweden is, he said, “enormous” for Saab.
Per a Saab slide shared by Nilsson, Sferical Al is built on Nvidia’s DGX SuperPOD architecture and is intended to support the development of Large Language Model and Vision‑Language Model capabilities to aid agent and operator communications. Saab quantified the computational leverage: when Sferical Al is applied to a one‑hour simulated training flight, it can produce the equivalent of one billion flight hours for a human pilot. “The exponential [computational] curve is … really hard to comprehend,” Nilsson explained.
How technologists, policymakers, and procurement leaders should respond
- Technologists and security teams: will need to evaluate how Sferical Al’s DGX SuperPOD‑based capacity is provisioned and how data control is enforced — Saab has tied the initiative explicitly to avoiding reliance on external cloud providers and preserving data governance within Sweden.
- Policymakers and regulators: should note that Saab framed Sferical Al as a government‑backed initiative; oversight and rules that govern domestic compute resources and classified data handling will become relevant as the Academy’s work scales into future programs.
- Procurement leaders and military program managers: are given a concrete development path to a potential loyal wingman capability — Saab said it intends to leverage its in‑house AI work to advance three unmanned aircraft concepts (the A1, A2 and A3) on the path toward a new loyal wingman drone.
Where Saab goes from here
Saab’s announcement ties three threads together: experimental autonomous agents on the Gripen E under BVR conditions, an institutional training and evaluation construct in the AI Fighter Pilot Academy, and a domestic supercomputing effort, Sferical Al, to sustain large‑model development and data control. The company explicitly links those technical investments to future air programs — including the A1, A2 and A3 unmanned designs and a loyal wingman concept — but also acknowledges the central challenge of translating flight‑test success into a “dependable capability that can learn, improve, and be relevant over time,” in Nilsson’s words.
The immediate, verifiable facts are straightforward: Saab has flown in‑house AI agents on a Gripen E, prioritized BVR scenarios, invested in a Swedish compute initiative built on Nvidia’s DGX SuperPOD, and signaled plans to apply the work to its upcoming unmanned‑aircraft line. The unanswered technical and programmatic questions — and how Sweden’s Sferical Al will be governed and used in practice — are the next chapters Saab has put in motion.
