DARPA ACE

DARPA’s Air Combat Evolution programme — ACE — set out in 2019 to answer a deceptively simple question: can a pilot trust an algorithm to fly the most demanding part of an air battle? The programme chose within-visual-range dogfighting, the close-in, high-stakes manoeuvring that pilots train for years to master, as its proving ground. The goal was not just to build a capable AI but to calibrate human trust in it, so that a pilot could focus on strategy while autonomy handled the tactical fight.

ACE first drew wide attention in August 2020 with the AlphaDogfight Trials. An AI agent built by Heron Systems beat seven rival teams’ algorithms and then defeated an experienced Air Force F-16 pilot five to nothing in simulated gun-only engagements. The result was a simulation with simplified flight physics, not real combat — but it was a striking demonstration. Among the competing performers were Shield AI , which would later acquire Heron Systems, and EpiSci , alongside Lockheed Martin, Aurora Flight Sciences, and others.

The harder test came in the air. ACE’s agents were loaded onto the X-62A VISTA, a heavily modified F-16 operated by the US Air Force Test Pilot School at Edwards Air Force Base. AI took control in live flight from December 2022, and in September 2023 the X-62A flew the first AI-versus-human dogfight against a crewed F-16 — a milestone announced in April 2024. In May 2024, Air Force Secretary Frank Kendall rode in its front seat during an AI-flown engagement and said afterwards that he would trust the system to employ weapons. Kendall has credited ACE’s progress with reinforcing his decision to press ahead with the Collaborative Combat Aircraft programme, the service’s effort to field autonomous drone “wingmen.”