An F-16 Fighting Falcon drawn from the U.S. Air Force's operational inventory has spent the past several weeks flying portions of its sorties with software, not the pilot strapped into the seat, at the controls. DARPA and the Air Force disclosed the milestone on July 16, describing flights out of Eglin Air Force Base, Florida, in which a test pilot performed the takeoff, handed the aircraft to an onboard artificial intelligence agent, then monitored while it flew — ready to take control back at any moment.
The distinction both organizations are drawing is not that an AI flew a fighter jet; that happened years ago on a bespoke research aircraft. It is that an AI flew this fighter jet — a standard fleet F-16, converted with a bolt-on kit, without rewriting the jet's core software.
Six jets, one switch
The aircraft is one of six F-16Cs being converted under the Viper Experimentation and Next-gen Operations Model – Autonomy Flying Testbed, or VENOM, a joint Air Force–DARPA effort spun out of DARPA's Air Combat Evolution (ACE) program. The first three jets arrived at Eglin in April 2024; the sixth landed April 1, 2025.
The modification package, the VENOM Autonomy Kit, was designed and integrated by multiple contractors under ACE. DARPA says it interfaces with the aircraft's existing flight controls and mission systems in a way that lets the pilot "toggle between traditional human control and AI control with the flip of a switch."
The pace has been deliberate. Ground testing ran for months, backed by simulation work dating to 2024. The jets flew in June 2026 purely to verify airworthiness with the new equipment aboard; only in July did the team hand control to an agent in flight.
"Getting the aircraft into the air is always a monumental milestone for a complex test program," said Tim Stevens, a VENOM test pilot with the 40th Flight Test Squadron at Eglin. "It represents years of design, modification and test planning poured into this project by a dedicated team of hundreds. As we cross this starting line, we are excited to watch VENOM redefine the boundaries of autonomous flight."
Brig. Gen. James "Fangs" Valpiani, the DARPA program manager who oversaw it, framed the engineering choice as the point. "The Air Force and DARPA team has automated flight controls and sensors on a standard F-16 without changing the jet's core software," he said. "This enables an efficient pipeline for developing dominant AI for aerial combat." Valpiani left DARPA in late July; Lt. Col. Patrick "Dice" Highland is the incoming manager.
What "combat testing" does and does not mean
Some coverage has described VENOM as moving into combat testing. That deserves precision: nothing here involves combat, weapons employment, or operational deployment. What is expanding is scenario complexity. VENOM aircraft are to be the platform for DARPA's Artificial Intelligence Reinforcements (AIR) program, which will run competing AI agents in live flight and scale toward multi-ship, beyond-visual-range engagements on a test range.
One widely repeated figure also needs correcting. Reports of an AI-controlled fighter flying 17 hours without pilot input are accurate, but do not describe VENOM. That milestone belongs to the X-62A VISTA, a one-of-a-kind modified F-16D, which logged more than 17 hours of AI-flown time across 12 test flights at Edwards Air Force Base between Dec. 1 and Dec. 16, 2022, and later flew the first AI-versus-human dogfight in April 2024. VENOM's own flight hours have not been disclosed.
Nor is the program expensive by fighter-development standards: roughly $50 million requested in fiscal 2024 and just under $17 million in fiscal 2025 — figures reflecting test infrastructure rather than procurement.
Analysis: the oversight question moves to the flight line
VENOM's real significance is as a feeder system. DARPA states plainly that capabilities matured under AIR will support Collaborative Combat Aircraft programs — uncrewed jets flying alongside crewed fighters. That effort has moved fast: General Atomics' YFQ-42A first flew in August 2025, Anduril's YFQ-44A on Oct. 31, 2025, both entered production under June 2026 contracts, and the first CCA live-fire test was announced in July. The service has signaled interest in 100 to 150 Increment 1 aircraft against a long-term ambition of roughly 1,000.
DARPA's language throughout is "human-on-the-loop" — a human supervising, able to intervene, but not approving each action. That maps onto DoD Directive 3000.09, issued in 2012 and updated in January 2023, which requires "appropriate levels of human judgment over the use of force" rather than a human in every decision. That gap is where the argument lives.
Proponents say this is how trust should be built: incrementally, on instrumented ranges, with a pilot holding a switch. Former Air Force Secretary Frank Kendall, at a Mitchell Institute roundtable in July, recalled a young fighter pilot's answer when he asked why there was so little resistance to CCAs: "Sir, these things are going to save our lives." In simulation, Kendall added, "the guys with the CCAs are beating the crap out of the guys without them, and not losing any manned fighters."
Critics — Human Rights Watch, the International Committee of the Red Cross, the Stop Killer Robots coalition — contend that "appropriate" is a word doing enormous work, and press instead for "meaningful human control." Their objection is about trajectory: each demonstration that autonomy is safe and cheap under supervision makes the supervision easier to relax later. Roboticist Noel Sharkey has argued a veto power is not meaningful control if the machine selects the target. Valpiani acknowledged the unresolved core of it, saying "many hard questions remain concerning the performance and trustworthiness of combat AI in the extreme fog and friction of modern warfare."
What to watch
Three things. First, whether the remaining VENOM jets reach flight status, and whether the Air Force releases cumulative autonomous flight hours — the figure that would let outsiders judge maturity independently. Second, how AIR structures multi-agent testing, and whether agents are evaluated against published criteria. Third, congressional posture: the fiscal 2026 NDAA authorized $900.6 billion in defense spending and requires reporting on AI-enabled systems, but added no restrictions on autonomous weapons beyond Directive 3000.09. If oversight tightens, it will likely arrive attached to CCA production, not to a research program flying six F-16s over the Florida panhandle.
“Many hard questions remain concerning the performance and trustworthiness of combat AI in the extreme fog and friction of modern warfare.”— Brig. Gen. James Valpiani, DARPA program manager, VENOM/AIR