The US is testing AI flight control on a standard F-16

From a special test aircraft to a standard platform

An AI agent has controlled a modified US Air Force F-16 in flight while a human pilot monitored the test from the cockpit.

The flight took place in July under a programme called VENOM, short for Viper Experimentation and Next-generation Operations Model — Autonomy Flying Testbed. The Air Force announced the milestone on Thursday, alongside DARPA, the US defence research agency.

The name is complicated. The basic change is easier to understand: a familiar military aircraft is becoming a reusable test platform for autonomy software.

That makes this more than another demonstration in a one-off research plane. It also does not make the F-16 an operational autonomous weapon. The public information stops well before that point.

DARPA and the Air Force have already tested AI-controlled flight in the X-62A VISTA, a heavily modified F-16 used for experimental work. Those earlier flights included close-range air-combat exercises between an AI-controlled aircraft and a human-piloted F-16.

VENOM is the next practical step. The Air Force has modified a group of operational-style F-16 testbeds with hardware, software and instrumentation that allow autonomy agents to interact with the aircraft’s flight and mission systems.

The new aircraft first flew in June to check that the modifications worked safely. The programme then moved to in-air autonomous control in July, according to the Air Force.

The agency has not published the route, length or manoeuvres from the autonomous portion of the flight. It describes the test as successful but provides no numerical result.

A pilot is still in the cockpit

The pilot does not disappear in this phase.

Official programme descriptions say the human can switch the aircraft between traditional and AI control. During the autonomous tests, the pilot remains in the cockpit to monitor the agent and make sure the flight and mission-system objectives are met.

This is often described as human-on-the-loop control. The system acts, while a person watches and can take over.

That arrangement reduces some immediate test risk. It does not answer the harder questions about future uncrewed aircraft, where a human may supervise several systems from somewhere else and may have less time or information to intervene.

VENOM is designed partly to study those later situations.

The next tests will be more crowded

The earlier Air Combat Evolution programme focused heavily on close-range, one-on-one flight. DARPA’s newer Artificial Intelligence Reinforcements programme is aimed at more complex conditions.

The agency says it wants to test multiple AI agents in live flight, including scenarios involving several aircraft and engagements beyond visual range. One eventual goal is to let a person command a group of autonomous uncrewed aircraft rather than directly fly each one.

That is where a flying testbed becomes useful. Researchers can install a new agent, place it in a real aircraft with real mission systems, observe how it behaves and change the software for the next flight.

Simulators remain important, but they cannot reproduce every sensor error, communication delay or physical constraint. A real F-16 makes those limits harder to ignore.

What the announcement does not show

The words AI-controlled F-16 are likely to travel further than the details.

The announcement does not say an AI system chose a target, fired a weapon or made an independent decision to use force. It does not publish evidence that the agent outperformed a human pilot. It does not give a reliability rate or explain how often the safety pilot intervened.

It also does not say when this technology might leave testing and enter operational service.

The distinction between flight control and the use of force is especially important. An agent can steer, manage speed or follow a tactical plan without being authorised to decide who or what may be attacked. Those decisions involve separate rules, command structures and legal obligations.

The official material offers little detail on those boundaries.

Why the test still matters

The milestone matters because it makes autonomy easier to test repeatedly on a relevant military platform.

The difficult part of deploying AI in aircraft is not only producing a strong model in a simulation. Engineers need to understand how it fails, how a person recognises that failure and how control moves back to the human without creating a new hazard.

As more agents and aircraft enter a scenario, those questions get harder. A supervisor may need to understand several automated decisions at once. Communications may be incomplete. The models may encounter conditions that were rare or absent during training.

The Air Force says VENOM is intended to help move autonomy from research into future uncrewed combat aircraft. That is an institutional goal, not a result delivered by this first flight.

For now, the clearest fact is modest but real: an AI agent has moved from a specialised experimental jet into the controls of a modified standard F-16 testbed, with a person still sitting close enough to take over.

The next useful announcement will need more than a photograph and a milestone. It should show how the agent performed, where it failed and what authority it was actually given.