This article is based on an official press release from Lockheed Martin.

Lockheed Martin and USAF Demonstrate Autonomous Missile Evasion on X-62A VISTA

On February 23, 2026, Lockheed Martin Skunk Works® and the U.S. Air Force Test Pilot School (TPS) announced a significant advancement in autonomous flight capabilities. Through a collaborative effort known as the “Have Remy” Test Management Project (TMP), the team successfully demonstrated the transfer of artificial intelligence (AI) agents from a digital simulation environment directly to the cockpit of the X-62A VISTA (Variable In-flight Simulator Test Aircraft).

According to the company’s announcement, this marks the first time a Lockheed Martin AI system has been granted direct control of the X-62A to perform complex tactical maneuvers specifically designed for autonomous missile evasion. The tests, conducted at Edwards Air Force Base in California, utilized a “sim-to-real” workflow that allowed engineers to rapidly update and refine the software during the flight campaign.

Project “Have Remy” and the X-62A

The X-62A VISTA, a highly modified F-16D Block 30, has become a central platform for the U.S. Air Force’s exploration of autonomous combat. While previous campaigns, such as the DARPA Air Combat Evolution (ACE) program, focused on within-visual-range “dogfighting,” the “Have Remy” project shifted the operational focus to survival.

Lockheed Martin reports that the campaign included over 100 test points where AI agents were placed in direct control of the aircraft. The primary objective was to validate that agents trained entirely in a digital environment could successfully execute split-second, 3D maneuvers to defeat incoming missile threats in the real world.

The “Supermassive” Simulation Engine

A key component of this success was the use of “Supermassive,” Lockheed Martin’s proprietary simulation environment. The company states that this system ran billions of simulated missions to train the AI agents on various threat scenarios before they were ever uploaded to the aircraft.

This digital foundation enabled a rapid “fly-fix-fly” development cycle. According to the press release, engineers were able to observe the AI’s performance in the air, identify issues, retrain the agents in the simulator, and push updated code back to the X-62A within hours. This stands in contrast to traditional software development cycles, which can often take weeks or months to implement flight-test feedback.

Workforce Development and Safety

The “Have Remy” project also served as a curriculum tool for the USAF Test Pilot School. Students and instructors were integrated into the development loop, defining the missile-evasion scenarios and evaluating how well the AI replicated its simulation performance in physical flight.

Safety remains a paramount concern in autonomous flight testing. The X-62A VISTA utilizes a unique safety architecture that allows a human safety pilot to remain in the cockpit. This pilot can instantly disengage the AI and take manual control if the system exceeds established safety limits. Lockheed Martin emphasized the importance of this human-machine teaming in their statement:

By proving that autonomous agents can be safely monitored, understood, and controlled in real time, we are turning the vision of human-machine teaming into an operational reality.

AirPro News Analysis

The shift from offensive dogfighting to defensive missile evasion represents a critical maturation in tactical AI. While dogfighting demonstrates the ability to outmaneuver an opponent, missile evasion requires the AI to prioritize survival against high-speed, non-cooperative threats, a fundamental requirement for any future autonomous wingman or unmanned combat aerial vehicle (UCAV).

Furthermore, the involvement of TPS students suggests the Air Force is actively working to build “trust” in non-deterministic systems. By training the next generation of test pilots to evaluate AI behaviors, the service is preparing its workforce for future platforms, such as the Next Generation Air Dominance (NGAD) family of systems, where human operators will likely manage teams of autonomous assets.

Sources

Sources: Lockheed Martin