Lockheed Martin says its Skunk Works team, the U.S. Air Force Test Pilot School and industry partners demonstrated sensor-driven AI autonomy aboard the X-62 VISTA: over eight flights, the aircraft completed 27 AI-controlled intercepts of a live T-38. The test is a flight demonstration—not evidence that an autonomous fighter capability is in operational service.
What Skunk Works demonstrated on the X-62
In an August 4, 2026 announcement, Lockheed Martin described a test intended to connect development, simulation, training and flight execution. The X-62 Variable In-flight Simulation Test Aircraft tracked a live T-38 using a Lockheed Martin Legion Pod. Secure data from that operational sensor was sent to an AI agent, which autonomously maneuvered the X-62 into a tactical intercept position. Lockheed Martin reported 27 AI-controlled intercepts across eight flights. Lockheed Martin’s announcement and its X-62 test account are the sources for these results.
As an Amazon Associate I earn from qualifying purchases.
The change emphasized by the company is the move from simulated target data to real-time streams from an onboard sensor. That makes the demonstration relevant to the sensor-to-action problem: an AI system must interpret incoming information and act on it during flight, rather than merely produce a response to a simulated input.
Lockheed Martin said full agent integration and ground testing with the X-62 took three months, and credited its “Supermassive” AI agent-generation capability with improving speed and agility. Those are company claims; the announcement does not publish a benchmark, independent test report or comparison with a baseline. It also does not specify the AI model, its detailed training procedure, scoring rules or sensor performance figures, so those details cannot be assessed from the announcement.
What the test does—and does not—establish
The reported event is a flight-test demonstration involving a piloted test aircraft modified for experimentation. It shows, according to Lockheed Martin, that an AI agent used live sensor information to fly the X-62 into an intercept position. It does not establish that an AI-controlled fighter is deployed, cleared for combat, or able to perform reliably across operational conditions. The company’s account is authoritative for what it says it tested, but it is not independent validation of performance.
Lockheed Martin executives framed the result as a step toward closing the sensor-to-action loop. Ron Fehlen, vice president and general manager of Lockheed Martin Skunk Works, said on August 4, 2026: “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.” Stacy Kubicek, vice president and general manager of Lockheed Martin Sensors and Global Sustainment, said the project demonstrates how sensing and autonomous AI can combine to support faster, more informed action in complex environments. These statements describe the company’s interpretation of the test, not independently published findings.
What Lockheed Martin says comes next
The company says the X-62 Mission Systems Upgrade will build on the experiment. Its stated goal is to support integration of combat systems, sensors and airborne AI agents in a next-generation mesh network. That is a planned next step, not a capability demonstrated in the August 2026 intercept flights. Lockheed Martin’s upgrade description outlines that intended direction.
The company’s broader capability page presents Skunk Works autonomy work across air and ground missions and identifies VISTA as the centerpiece of DARPA’s Air Combat Evolution program. That is Lockheed Martin’s program description, not a separate evaluation of the X-62 test. Lockheed Martin’s autonomy overview provides that context.
Other Skunk Works AI projects are separate efforts
The X-62 intercept is one part of a wider set of autonomy projects. They address different tasks and have different levels of demonstrated maturity; they should not be treated as components of the same system or as equivalent evidence of autonomous combat capability.
| Project | Task and input | Demonstrated human or system role | Evidence described |
|---|---|---|---|
| ISR anomaly detection with Arquimea | In an April 2025 simulated small-UAS scenario, neural-network memories analyzed EO/IR imagery to identify changes in physical features from new viewpoints. | Detection research; the release does not describe autonomous aircraft maneuvering. | Lockheed Martin and Arquimea described research and development and said they would explore applications to other sensors and autonomous-system decisions. April 2025 announcement |
| Mission-contingency replanning with Stalker XE Block 25 and Alta X 2.0 | A December 2025 demonstration simulated fuel contingencies in a UAV mission. | A ground C2 system generated replanning options; an operator selected an option, after which AI reassigned tasks and directed the Stalker to return to base. | Lockheed Martin reported a live demonstration using an Alta X 2.0 drone modified by Drone Amplified. December 2025 announcement |
| Indago 4 swarm operations with Red Hat | Autonomous swarm operations using the Indago 4 tactical quadcopter. | The feature describes Red Hat Device Edge combined with Skunk Works swarm operations; it does not give a comparable operator-action breakdown. | Lockheed Martin’s 2025 feature describes the collaboration. 2025 feature |
These examples show a range of work—from simulated image analysis, to operator-approved mission changes, to flight-test autonomy. The sources describe company demonstrations and program material; they do not establish independent operational performance across these efforts.
Rank #4
How to read the modernization claim
For readers evaluating what “integrating AI” means here, the useful distinctions are the input, task, human role and test maturity. The X-62 demonstration used a live sensor feed for an AI-controlled maneuver during flight testing. The contingency exercise involved a human choosing among options before the system acted. The anomaly-detection work was described in a simulated scenario. A planned mesh-network upgrade is a future integration objective, not a completed demonstration.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
That is meaningful progress in testing how AI can connect sensing with action, but the available company accounts leave key questions unanswered: how performance was scored, how the system handles uncertainty or unexpected inputs, and how it compares with human or non-AI baselines. Without independent evaluation and operational evidence, the result is best understood as a reported test milestone in defense modernization—not proof of fielded autonomous fighter capability.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
