Skunk Works Just Flew an AI That Outperformed Its Simulation in Real Air Combat
Lockheed Martin's Skunk Works successfully flight-tested a sensor-powered AI that flew a real jet in air-to-air intercepts, performing better in live flight than in simulation. The program now extends into missile-evasion and human-machine teaming with the U.S. Air Force, laying groundwork for future Collaborative Combat Aircraft.
In June 2024, a Czech-made L-29 Delfin jet took off from Iowa City, Iowa, with an AI at the controls—not a human pilot. The AI, developed by Lockheed Martin's Skunk Works, executed eight tactical test cases per flight, including head-to-head engagements and missile-evasion maneuvers, against a virtual adversary. According to Dr. Tom "Mach" Schnell of the University of Iowa's Operator Performance Laboratory, the system "performed even better in live flight than in simulation." This was not a lab experiment or a drone test; it was a full-scale, sensor-powered AI flying a real jet in air combat scenarios.
What Happened: The AI That Flew a Jet in Real Combat Scenarios
The test, announced on June 5, 2024, was part of Skunk Works' Enhanced Collaborative High-Frequency Orientation System (ECHOS) program, which develops AI for air-to-air missions. The AI directly issued heading, speed, and altitude commands to the L-29 through a new flight interface, executing simulated-to-real transfer of behaviors learned in simulation. Each flight included eight test cases covering standard head-to-head engagements, off-aspect encounters (non-head-on geometry), missile support scenarios, and missile defeat (evasion) scenarios. The AI was tested in both offensive and defensive risk postures, and Lockheed Martin reported "clean sim-to-real transfer" with the AI agent appearing "intentional and decisive."
💡 The AI performed better in live flight than in simulation—a rare and significant result that suggests the training environment was conservative and the AI generalized well to real-world dynamics.
The L-29 was supplied by the University of Iowa's Operator Performance Laboratory (OPL), which partnered with Skunk Works on the integration. The test also involved a virtual adversary, meaning the AI was not fighting another real aircraft but a simulated opponent—a common approach to reduce risk while validating the AI's decision-making in real flight conditions.
Why It Matters: From Simulation to Supersonic Missile Evasion
This test is part of a broader Skunk Works autonomy roadmap that includes missile-evading AI on the X-62A VISTA, an F-16-based testbed. In earlier tests, Skunk Works demonstrated tactical AI that took control of the X-62A to evade a simulated incoming missile at supersonic speeds up to ~1,200 mph (Mach 1.6). That work directly relates to the "missile defeat" scenarios exercised in the L-29 intercept tests.
Lockheed Martin's O.J. Sanchez, vice president and general manager, referenced the X-62A tests in Aviation Week coverage. The company is also working with the U.S. Air Force Test Pilot School (TPS) on human-machine teaming, where student test pilots flew with AI in missile-evasion scenarios requiring split-second decision-making. A Lockheed Martin YouTube video describes the AI as "transparent, controllable and built for real missions," emphasizing that the goal is human-machine teaming, not fully autonomous operation.
💡 Skunk Works is deliberately designing AI to be a teammate, not a replacement—a critical distinction for military acceptance and safety certification.
Additionally, a Lockheed Martin Instagram reel showed an F-22 controlling an uncrewed system directly from the cockpit, demonstrating the command-and-control architecture needed for future AI-enabled fighter-drone teaming. These tests collectively support the U.S. Air Force's Collaborative Combat Aircraft (CCA) program, where unmanned combat aircraft will accompany manned fighters.
What It Means for Business: The Path to CCA and Beyond
For defense contractors and technology companies, this test validates that AI can be trained in simulation and deployed on real aircraft with minimal degradation—a key milestone for the CCA program. Skunk Works uses open mission systems standards to ensure broad compatibility and rapid transition to future platforms, meaning the AI architecture could be adapted for multiple aircraft types, not just Lockheed Martin platforms.
The practical implications: AI that can perform air-to-air intercepts and missile evasion could reduce pilot reaction times, enable maneuvers beyond human G-limits, and allow a single pilot to command multiple unmanned combat aircraft. However, TURDEF notes that such AI-driven fighter autonomy is not expected to be fielded in the short term due to ongoing development, training, and exploration requirements.
💡 For startups and investors, the key opportunity is not in building the AI itself but in developing the data pipelines, simulation environments, and integration tools that make sim-to-real transfer work reliably. Lockheed Martin's success with ECHOS suggests that the bottleneck is shifting from algorithms to data curation and test infrastructure.
Lockheed Martin has scheduled several additional flight tests for 2024 and beyond, increasing complexity by introducing additional aircraft, exploring offensive counter-air missions, and testing battle management scenarios where AI supports broader tactical decision-making.
What to Watch Next
Watch for Lockheed Martin's next flight tests, which will include multiple aircraft in the same scenario—a significant step toward realistic teaming. Also monitor the X-62A VISTA program, where tactical AI is being tested in increasingly aggressive maneuvers. If the AI continues to outperform simulation, it could accelerate the timeline for CCA deployment. Finally, pay attention to how the U.S. Air Force Test Pilot School's human-machine teaming results feed into operational requirements—the AI's "transparent and controllable" design philosophy may become a template for military AI development across the services.
Want automation like this for your business?
Get in touch and we'll show you exactly what's possible for your setup.