This article is published in Aviation Week & Space Technology and is free to read until Oct 09, 2026. If you want to read more articles from this publication, please click the link to subscribe.
Human pilots will play the role of safety observers in testing of new AI algorithms designed for beyond-visual-range air combat.
For years, advanced artificial intelligence agents designed to outthink adversaries in dogfights have thrived in the consequence-free realm of simulators, but the gap between digital models and reality is starting to close.
The arrival in June of a newly modified fleet of Lockheed Martin F-16C Block 42 fighters marks a critical expansion in tactical aviation infrastructure. Over at least the next two years, this cadre jointly funded by DARPA and the U.S. Air Force will be dedicated to exposing artificial intelligence (AI) agents to the reality of flight-testing beyond-visual-range combat between multiple aircraft for the first time.
- VENOM-equipped F-16s slated to start autonomy testing next year
- The VENOM Autonomy Kit continues testing for safety of flight
“These are combat-capable jets that have fully live sensors on them to allow us to truly vet out these types of scenarios and not just simulate everything that we have on there,” Terry Wilson, deputy manager for DARPA’s AI Reinforcements (AIR) and Viper Experimentation and Next-generation Operations Model (VENOM) testbed aircraft, tells Aviation Week.
AI-enabled air combat is close to operational reality, regardless of whether the algorithms are ready. The first increment of the Air Force’s Collaborative Combat Aircraft (CCA) program entered production only a few weeks after the first three VENOM-configured F-16s entered safety-of-flight testing in early June at Eglin AFB, Florida. The first General Atomics Aeronautical Systems Inc. FQ-42 and Anduril FQ-44 jets could arrive at Test and Training Unit-1 (TT-1) at Creech AFB, Nevada, in two years.
The first production CCAs stand to benefit from nearly a decade of flight testing. In addition to the 12 prototype aircraft assigned to the Air Force’s Experimental Operations Unit (EOU) at Creech, DARPA’s infrastructure for AI-enabled flight tests began with another heavily modified F-16 operated by the Test Pilot School at Edwards AFB, California. This ex-Israeli Air Force F-16—designated since 2021 as the X-62 Variable In-Flight Simulator Aircraft (VISTA)—played a pioneering role in the transition of autonomy from software pipelines to real aircraft.
The X-62 became the first aircraft to fly with AI agents under DARPA’s Air Combat Evolution (ACE) program. Following a heated competition between multiple companies, an AI agent created by Heron Systems—now owned by Shield AI—defeated a human F-16 pilot in a simulator on the ground five times in a row to win DARPA’s AlphaDogfight trials. In the next phase of the ACE program, DARPA migrated more than a dozen AI agents to flight testing on the X-62.
The program helped transition AI agents from simulators to flight demonstrations, but there were limits. The Test Pilot School is now upgrading the X-62 with the RTX Phantom Strike radar, but the aircraft lacked combat-rated mission systems at the time of the ACE program. DARPA also focused the demonstrations on dogfights within visual range between only two aircraft. As autonomy in combat aviation continues to develop, future AI agents will need to be able to process live data from onboard sensors for a multitude of beyond-visual-range targets.
“We went from doing within-visual-range-type combats under ACE to beyond-visual-range under AIR,” Wilson says. “The other thing is that it brings live sensors. So these are combat-capable jets that have fully live sensors on them to allow us to truly vet out these types of scenarios and not just simulate everything.”
DARPA and Air Force officials declined requests to identify the mission systems onboard the VENOM fleet, saying those details are not releasable. Because they are underpowered by Pratt & Whitney F100-220 engines, the VENOM-configured F-16s are not currently combat-deployable. They operated as test aircraft at Nellis AFB, Nevada, before arriving at Eglin in April 2024.
The Air Force spent two years modifying the F-16s with the VENOM Autonomy Kit (VAK). Similar to the System for the Autonomous Control of Simulation (SACS) installed on the X-62, the VAK establishes a set of parameters for AI agents to take control of the jets in flight safely. The presence of a human pilot onboard the VENOM aircraft requires the VAK to establish what Wilson calls a “safety sandbox” that limits the authority of the AI agents.
The VAK was necessary so “that these autonomous agents could fly them in a way that obviously wouldn’t harm the humans and wouldn’t break the jet,” Wilson says, noting that the VAK represents an improvement over the original SACS. “So we built that sandbox in VENOM, and that’s what you’re seeing as a culmination of that design.”
The Air Force testers are still proving that the VAK works as intended. The VENOM jets continue to fly only sorties focused on certifying the safety of the hardware and software. The goal is to give the AI agents the freedom to make tactical decisions but eliminate any commands that do not translate safely from a simulator environment to a real aircraft in flight.
“Integrating software is difficult, period,” Wilson says. “Integrating autonomy into platforms is also an added set of complexity. So flight schedules are subject to change. But I will say that over the next year, we’re looking at getting ourselves into these multi-ship tactical combat situations with autonomy flying these jets.”
The exact timing for autonomous flight tests has not been released, but the AIR program is funded through the end of fiscal 2028, a DARPA spokesperson says.
How the testing will feed into AI agent development for operational aircraft assigned to the TT-1 and EOU also is not clear. Part of the ambiguity is driven by differing architectural approaches between the operational CCAs and DARPA’s AIR programs. The operational CCAs are mandated to comply with the joint Air Force-Navy Autonomy Government Reference Architecture (A-GRA), which established standards for modularity and interfaces between hardware and software components. The modeling infrastructure and VAK suite were not initially designed to be compliant with A-GRA standards, but that could change.
“We are building an A-GRA compatibility or compliance into our modeling system environment and also with integration onto the airframe,” Wilson says. “How much of that or which parts are going to be fully implemented is not completely finalized yet.”
While the VAK-equipped, VENOM-branded F-16s offer visibility into the AIR program, an unseen effort is building the infrastructure for a powerful development pipeline of advanced autonomy capabilities. That infrastructure of highly accurate digital models is expected to produce new code on a timeline “orders of magnitude faster than the present state of the art,” DARPA budget documents state. Those AI-driven capabilities can then be verified in flight with the human-in-the-loop F-16 test fleet.
If some or all of that digital system becomes A-GRA-compliant, DARPA sees another potential benefit for future operational aircraft. The A-GRA standard is expected to evolve over time, and the AIR infrastructure for digital modeling can help inform those decisions.
DARPA is looking at “exploring those open environments and open standards to maybe provide feedback to the folks that are developing that on where things work well, where they don’t,” Wilson says.
Related Content
