Skunk Works Pushes AI Fighter Intercepts Into the Real World

The idea of an artificial intelligence system controlling a fighter aircraft in combat once belonged firmly in the realm of science fiction. Today, it is becoming an increasingly practical engineering challenge, and one that the aerospace and defence industry is tackling at remarkable speed. In the latest demonstration, Lockheed Martin Skunk Works®, working with the U.S. Air Force Test Pilot School (TPS) and industry partners, has taken another significant step towards autonomous air combat. During a series of flight tests at Edwards Air Force Base in California, an AI agent used information generated by an operational airborne sensor to autonomously execute air intercepts against a live aircraft. Across eight flights, the X-62 Variable In-flight Simulation Test Aircraft (VISTA) completed 27 AI-controlled intercepts, demonstrating that artificial intelligence can move beyond simulated targets and interact with the real-world sensor environment of a flying combat aircraft.

For years, much of the development surrounding autonomous combat aircraft has taken place in laboratories, simulators and carefully controlled test environments. Computers can process enormous quantities of simulated information and make decisions at speeds that humans cannot match, but the real challenge comes when those algorithms have to deal with the imperfect, rapidly changing information generated by actual sensors in a flying aircraft. The latest X-62 trials were designed specifically to address that challenge by demonstrating what Lockheed Martin describes as a complete "sensor-to-action" loop. The aircraft was equipped with the Lockheed Martin Legion Pod®, which provided sensor information while tracking a live T-38 aircraft. That information was securely passed to an AI agent, which interpreted the sensor data and autonomously manoeuvred the X-62 into a tactical intercept position. In other words, the AI was not simply following a predetermined flight path or responding to a computer-generated target. It was receiving information from a real airborne sensor, processing that information and using it to determine how the aircraft should respond.

"Our ongoing partnership with TPS is driving important progress with this latest flight test series demonstrating that our AI can effectively and reliably close the sensor-to-action loop aboard an operational combat aircraft," said Ron Fehlen, vice president and general manager of Lockheed Martin Skunk Works. According to Fehlen, the autonomous agents consumed classified infrared search and track feeds and executed combat-critical manoeuvres in real time. The achievement is significant because modern fighter aircraft already generate vast quantities of information through radar, infrared sensors, electronic warfare systems, datalinks and other mission equipment. The challenge is no longer simply collecting information, but processing it quickly enough to provide a tactical advantage.

AI as a combat force multiplier is where artificial intelligence could become an important force. A human pilot can process an extraordinary amount of information, particularly with the assistance of modern avionics and sensor-fusion systems, but there are unavoidable limits to human reaction time, attention and workload. In a high-speed air engagement, where several aircraft may be manoeuvring simultaneously, and the tactical situation can change in seconds, those limitations become increasingly important. An AI system can continuously process sensor information and respond within extremely short timeframes. Properly designed and tested, such systems could take over some of the more demanding or time-critical tasks, allowing the pilot to concentrate on the wider tactical situation and decisions that require human judgement.

Stacy Kubicek, vice president and general manager of Lockheed Martin Sensors and Global Sustainment, highlighted the importance of connecting sensing technology directly to autonomous decision-making. "Our ability to provide reliable sensor data is critical, but the real advantage comes when that data can connect seamlessly with AI to take action," she said. The X-62 demonstration is therefore not simply about teaching an aircraft to fly itself. It is about establishing a practical connection between sensors, mission systems and autonomous decision-making. This distinction will become increasingly important as future combat aircraft operate as part of larger networks, receiving information from multiple platforms and sensors and using artificial intelligence to help make sense of the rapidly changing battlespace.

One of the more impressive aspects of the programme is the speed at which the technology was moved into the aircraft. Lockheed Martin says its "Supermassive" AI agent generation capability significantly improved development speed, with full integration and ground testing of the agents with the X-62 completed in approximately three months. That rapid development cycle is important because the future of military aviation is likely to involve rapidly evolving software as much as hardware. Traditional combat aircraft programmes can take years or even decades to move from concept to operational capability, whereas software-based capabilities can potentially be developed and updated much faster, provided they can be tested and validated to the demanding standards required for military aviation.

The X-62 VISTA provides an ideal environment for this work. Based on a heavily modified F-16D, the aircraft has been used for decades as a flying laboratory for advanced flight-control technologies, autonomy research and other experimental systems. Rather than waiting for an entirely new combat aircraft to be developed, engineers and test pilots can use the X-62 to introduce new software, sensors and mission systems and evaluate their performance in realistic flight conditions. This makes VISTA an important bridge between laboratory development and operational aviation. It also allows engineers to discover problems that may not appear during simulation. An algorithm may perform perfectly when operating with clean and predictable data inside a computer, but the real world is considerably less cooperative, with sensor noise, changing target behaviour, aircraft limitations and other variables constantly influencing the aircraft.

While the phrase "AI-controlled fighter" is guaranteed to attract attention, the current direction of development is more nuanced than simply removing humans from the cockpit. Much of the immediate focus is on AI augmentation rather than eliminating the pilot. A future fighter pilot may find that the aircraft is capable of performing many tasks that previously demanded constant attention, including elements of sensor management, threat prioritisation and tactical manoeuvring. This could give the pilot more time and mental capacity to concentrate on the broader tactical picture and decisions that require human judgement.

That could also have direct implications for pilot survivability. Modern combat aircraft are incredibly complex, and pilots can become overloaded when required to simultaneously manage aircraft systems, sensors, communications, weapons and tactical information. Delegating some of those tasks to autonomous systems could reduce workload and allow the pilot to concentrate on the most important decisions. At the same time, giving an AI system control of a fighter introduces major questions surrounding reliability, cybersecurity, human oversight and rules of engagement. The technology must therefore prove itself not only capable, but predictable, reliable and controllable.

The latest X-62 flights are part of a much larger development effort. Skunk Works has been involved with the X-62 programme for decades and has helped provide open software and hardware architectures designed to support advanced flight testing. The next stage will see the aircraft's Mission Systems Upgrade used to demonstrate more comprehensive integration between combat systems, sensors and airborne AI agents. The objective is to create a more seamless environment in which autonomous systems form part of a wider network connecting aircraft, sensors, weapons and other platforms.

This concept is becoming increasingly important as air forces look towards networked combat operations. An aircraft could potentially detect a target using one sensor, receive additional information from another aircraft or platform, process the combined information through an AI system and then determine the most appropriate tactical response. The aircraft itself becomes one element of a much larger combat network. Open architectures will be particularly important in this environment, allowing new sensors, algorithms and mission systems to be integrated as they become available rather than locking future capabilities to a single generation of hardware.

The significance of the latest demonstration is therefore considerably greater than 27 successful intercepts. The trials show that AI can be connected to real airborne sensor information and used to control a fighter aircraft during live flight operations, moving autonomous technology another step away from the laboratory and towards genuine operational relevance. For the U.S. Air Force, the partnership with the Test Pilot School provides an opportunity to evaluate increasingly autonomous aircraft under realistic conditions, while for industry it offers a pathway to rapidly develop and refine technologies that could eventually find their way into future combat platforms.

For pilots, this could result in a very different relationship between human and machine. The fighter of the future may still have a cockpit, wings and an engine, but much of its combat capability could increasingly reside in software. Instead of manually controlling every aspect of the aircraft while attempting to interpret an overwhelming amount of information, the pilot could become the commander of a highly capable autonomous system. The aircraft would handle many of the lower-level tasks while the human remains responsible for mission-level decisions and oversight.

The X-62's latest flights provide a glimpse of what that future could look like. Artificial intelligence is no longer being tested solely against simulated targets in a laboratory. It is being connected to real sensors, placed aboard a real fighter aircraft and allowed to make decisions in real time. The machines are learning to fly the fight and the next challenge for the aviation industry will be deciding exactly how much of that fight humans should allow them to control.