The Air Force Can Flip a Switch and Hand an F-16 to an AI. It Did.

On a June morning at Eglin Air Force Base in Florida, a modified F-16 fighter jet lifted off with two entities at the controls: a U.S. Air Force safety pilot in the cockpit, and an AI agent running on hardware connected to the aircraft's systems. During the mission, the pilot flipped a switch. The AI took over.

That switch is the core of what DARPA calls the VENOM Autonomy Kit, or VAK. It is a hardware-and-software package intended to let an operational F-16 be autonomously piloted by an AI agent without changing the jet's core flight software. The pilot can hand control to the system mid-flight and take it back when needed.

This is the VENOM program — Viper Experimentation and Next-gen Operations Model — and its flight milestone is a practical step in how autonomous combat-aircraft technology is tested: on a real aircraft, with a human safety pilot still responsible for the mission.

From Research Aircraft to a Fleet-Relevant Testbed

DARPA's earlier Air Combat Evolution program used the X-62A VISTA research aircraft for AI dogfight experiments. Those tests demonstrated important capabilities, but VISTA is a purpose-built experimental platform rather than a jet from an operational flight line.

VENOM is designed around a different deployment question. Its kit connects to existing flight controls and mission systems through a hardware layer while leaving the F-16's core flight-control software intact. That approach could make it easier to test and iterate on autonomy across a familiar aircraft platform, though it does not remove the validation and certification work required before any operational use.

The program first flew piloted validation missions to establish that the installation was airworthy. It then tested the handoff to the AI agent while the safety pilot remained ready to retake control. Brig. Gen. James Valpiani described the result as infrastructure for developing trusted autonomous air-combat capabilities — an important distinction from a finished weapon or deployed combat system.

What the AI Did — and Did Not Do

The test did not demonstrate a complete autonomous combat mission. DARPA has said the agent controlled portions of flight, while specific maneuvers remain undisclosed for operational-security reasons. A pilot remained aboard throughout and could resume control through the same physical handoff.

That makes this human-on-the-loop testing, not human-out-of-the-loop combat. The distinction is consequential technically, legally, and ethically: the architecture gives the human supervisor a direct means to interrupt the system rather than treating oversight as a purely procedural promise.

Autonomous flight in managed conditions is not the hardest problem ahead. The harder question is tactical decision-making in adversarial conditions — when sensors are jammed, positioning signals are spoofed, and an opponent is deliberately trying to induce an error. VENOM's value is a way to collect and validate real-flight evidence against that harder trust problem over time.

The Next Test Is a Team, Not a Solo Flight

DARPA is planning multi-agent live-flight testing: multiple autonomous aircraft operating together, with people supervising at the mission level while agents manage portions of tactical execution. That is the point at which this becomes more than an advanced autopilot experiment.

AI-piloted aircraft may eventually maneuver beyond human physiological limits and take risks that crews cannot. But VENOM remains a test program, not an imminent operational deployment. It still needs validation across flight conditions and adversarial scenarios, while the policy framework for autonomous lethal systems remains contested.

The milestone is nevertheless concrete: the technical infrastructure is being built through controlled, reversible handoffs at real altitude, on a real aircraft. The safety pilot did not hand control to an AI and walk away; they stayed in the seat, monitored the aircraft, and retained the ability to take it back.

A Pattern for Safety-Critical AI

The broader lesson is not unique to military aviation. Autonomous vehicles, surgical robotics, and industrial automation all face a version of the same design challenge: when should an AI operate, how is it supervised, and how quickly can a human reliably override it?

Trust in those settings cannot be a legal formality. It needs to be earned with evidence about system behavior under novel and stressful conditions. VENOM is an example of the incremental approach: make the handoff bounded, keep a human close to the controls, and expand the operating envelope only when the evidence supports it.

Sources

DARPA: https://www.darpa.mil/news/2026/darpa-us-air-force-fly-ai-controlled-f-16

The Debrief: https://thedebrief.org/after-surviving-a-dogfight-in-a-test-aircraft-darpas-venom-ai-controlled-pilot-just-flew-a-modified-combat-style-f-16/

Army Recognition: https://www.armyrecognition.com/news/aerospace-news/2026/u-s-air-force-f-16-fighter-flies-under-ai-control-as-darpa-expands-venom-combat-tests

Stars and Stripes: https://www.stripes.com/branches/air_force/2026-07-20/first-venom-autonomous-jet-flight-22317435.html

Northeast Times: https://northeasttimes.com/2026/07/24/ai-flew-an-f-16-fighter-jet-here-s-what-actually-happened/