An AI agent flew a real F-16 fighter jet under DARPA's VENOM program
DARPA and the U.S. Air Force disclosed that an F-16 modified under the VENOM program (Viper Experimentation and Next-generation Operations Model) has flown with an onboard AI agent controlling portions of the flight, at Eglin Air Force Base.
How they got there
The program didn't jump straight to autonomous flight. It started with piloted flights in the modified aircraft to confirm the hardware and software upgrades were safe in the air at all, before any AI system touched the controls. Only after that validation phase did the program move to flights where an AI agent actively flew portions of the mission — with a human safety pilot in the cockpit the entire time, monitoring performance and able to intervene.
Why the staging matters
That sequencing is the actual story, more than the flight itself. "AI flies a fighter jet" is a striking headline; "AI flies a fighter jet after a validation phase specifically designed to catch problems before autonomy is ever engaged, with a human able to take over at any moment" is a much more boring — and more trustworthy — description of the same event.
The pattern here — validate the platform first, add autonomy incrementally, keep a human in the loop with override authority, expand scope only after each stage holds up — is the same playbook worth applying to autonomous systems in far lower-stakes domains too. If it's the right way to trust AI with a fighter jet, it's a reasonable bar for trusting it with your production deployment pipeline.
VENOM is explicitly a test and evaluation program, not a deployment — the goal right now is generating data on how AI agents perform in a real airframe under real flight conditions, not fielding autonomous combat aircraft. Worth watching as one of the more concrete, verifiable data points in the broader "how much do we actually trust AI with physical control" conversation.
Source: www.darpa.mil