The F-35 already uses advanced automation, and is now entering a true AI phase. Capabilities, limitations, missions, and exports to allies: what is changing.
In summary
The debate over the AI capabilities of the F-35 suffers from a simple problem: many comments confuse artificial intelligence, advanced automation, data fusion, and future software components still in testing. Today, the F-35 is not an “autonomous” fighter. It does not decide to fire on its own. However, it is already a formidable tactical processing machine. Its strength lies in sensor fusion, threat prioritization, track correlation, synthetic display on the helmet, and secure information sharing with other platforms. With Tech Refresh 3 and Block 4, this logic is changing on a whole new scale: more computing power, more memory, better target recognition, increased electronic warfare capabilities, and a technical base more conducive to the integration of AI models. Since February 2026, Lockheed Martin has even tested an AI Combat ID capability in flight on the F-35. This is significant. But we must remain precise: it is a promising tactical demonstration, not yet a widespread revolution across the entire fleet.
The real issue: the F-35 is not primarily a “smart” aircraft, it is an aircraft that merges
When we talk about F-35 artificial intelligence, we must first put the words in their proper place. The basis of the F-35 today is not decision-making AI in the popular sense of the term. Its operational core is based on an architecture of sensors, fusion, and restitution. The aircraft collects radar, infrared, electromagnetic, and data link signals, correlates them, and then presents the pilot with a synthetic image of the battlefield. It’s not “magic.” It is a very high level of automation, designed to reduce the pilot’s cognitive load and speed up human decision-making.
The US Air Force points out that the F-35 features the Distributed Aperture System (DAS), which provides spherical surveillance around the aircraft for missile warning, air warning, and day/night vision. Added to this is the EOTS for air-to-ground detection and targeting, as well as long-range air-to-air detection. The pilot’s helmet then concentrates this data on the visor. In other words, the F-35 doesn’t just give the pilot sensors. It saves him from wasting time mentally reconstructing the tactical situation. This is where its superiority begins.
This is also why we need to be honest: much of what many people call “AI” on the F-35 is, in reality, information fusion, automatic sorting, and decision support.
This is already considerable. In modern combat, reducing a few seconds of mental latency is often worth more than a marginal gain in speed. Lockheed Martin emphasizes the F-35’s ability to “collect, analyze, and share” data more than its complete software autonomy.
The technical foundation that makes AI credible on board
The real leap forward comes from Tech Refresh 3 (TR-3). This is the new hardware and software foundation that will enable the F-35 to support the developments of Block 4. The JPO clearly explains that TR-3 provides the computing power needed for modernized sensors, longer-range weapons, improved electronic warfare, more powerful data fusion, and increased interoperability between platforms. In simple terms: without TR-3, the F-35 remains a very good aircraft; with TR-3, it can accommodate more demanding, faster, and more ambitious functions.
Lockheed Martin describes TR-3 as an evolution of the mission architecture with a new integrated central processor, improved panoramic display, more memory, and other classified capabilities. This is essential to the topic of AI. Onboard AI capability does not depend solely on an algorithm. It depends mainly on three things: available power, speed of data access, and the ability to update the software without breaking the rest of the system. This is precisely what TR-3 seeks to improve.
The problem, and it must be said bluntly, is that this modernization is still behind schedule. In September 2025, the GAO reported that Block 4 had already cost more than $6 billion extra and that its completion was delayed by at least five years compared to initial estimates. The same report stated that TR-3, valued at $1.9 billion, was the main factor in the 2024 delivery delays. It is therefore not enough to promise AI. The software and hardware infrastructure that makes it usable must also be delivered.
The first real AI demonstration tested in flight
The most significant and recent development dates back to February 2026. Lockheed Martin announced that it had flight-tested an AI-enhanced Combat Identification capability integrated into the F-35’s information fusion system. The important point is not marketing. The important point is that the company claims that a tactical model has, for the first time in flight, generated independent combat identification, displayed directly to the pilot. This changes the nature of the debate: we are no longer talking about traditional automation, but rather an AI component applied to a specific tactical task.
In this test, called Project Overwatch, the AI/ML model was used to resolve ambiguities between different transmitters. In short, the system helped to more quickly distinguish the nature of signals or threats that, in a saturated environment, can appear similar. Lockheed adds that engineers were able to label new transmitters, retrain the model in a matter of minutes, and then reload the updated version for the next flight during the same mission preparation cycle.
This is probably the most interesting part of the announcement: the speed of tactical reprogramming. AI that is useful in warfare is not just high-performance AI. It is AI that can be quickly adapted to the situation.
However, we must keep a cool head. This demonstration does not mean that the entire F-35 fleet already has mature, validated, and widespread combat AI. Lockheed refers to flight results that will “inform” future developments and possible integration paths. This vocabulary is important. It indicates real progress, but still in transition between demonstration, qualification, and eventual production. Those who are already presenting this as a generalized operational standard are jumping the gun.
What AI can actually do for F-35 missions
The first obvious application is threat identification in a cluttered environment. In dense electronic warfare, a pilot does not lack information; he has too much. AI can then be used to classify, prioritize, and correct ambiguities between radar emissions, electromagnetic signatures, or enemy platform behaviors more quickly. This is exactly the use case for Project Overwatch. For a stealth fighter jet that is supposed to be the first to enter a contested anti-aircraft bubble, this capability is more than just a convenience. It is a survival multiplier.
The second application concerns target recognition. Lockheed Martin already associates Block 4 with improved target recognition. This can cover several realities: better multi-sensor correlation, more reliable discrimination between friendly, neutral, and hostile contacts, or even a reduction in false positives in saturated tactical scenes. Again, we must not fantasize about an aircraft that “chooses” its targets on its own. The real value is more sober, but very concrete: helping the pilot make decisions faster and with fewer errors.
The third application is non-kinetic electronic warfare. Block 4 is set to add advanced EW capabilities, and the JPO explicitly mentions improved functions in this area. Embedded or semi-embedded AI could eventually help to better recognize emission patterns, prioritize threats, recommend responses, and speed up the reconfiguration of threat libraries. This is a logical area, as modern electronic warfare is a problem of rapid signal analysis, and therefore a natural field for well-trained learning models.
The fourth application concerns the role of the F-35 as a combat node. The F-35 is not just a shooter. Lockheed presents it as a “force multiplier” capable of sharing information in real time between domains and between allied forces. In this logic, AI can be used not to “fly” the aircraft, but to better filter what needs to be disseminated, to whom, and in what order. In the era of distributed combat, the volume of information quickly becomes unmanageable. An aircraft that knows how to prioritize useful data can be more valuable than an aircraft that simply generates it.
Finally, there is the prospect of piloted and drone teaming. The F-35 website now highlights the tandem between piloted aircraft and drones. It is important to read this correctly: the F-35 is not being promoted as an autonomous drone, but as a platform capable of operating in an ecosystem where AI, distributed sensors, and unmanned devices work together. Here again, the onboard AI is mainly used to manage information, speed up tactical decisions, and coordinate the actions of a set of sensors and effectors more precisely.
The performance that really matters when talking about AI
On paper, the F-35A reaches Mach 1.6, or approximately 1,930 km/h (1,200 mph). Its combat radius on internal fuel is greater than 1,093 km (590 nmi) and its internal range exceeds 2,200 km (1,200 nmi). It can carry up to 8,160 kg (18,000 lb) of payload. These figures are impressive, but they don’t tell the whole story when it comes to AI. The F-35’s true performance is no longer just kinematic. It is cognitive. The aircraft wins because it sees, merges, prioritizes, and shares before the other does.
This is precisely what explains the logic behind the program. As of February 2026, the F-35 had more than 1,300 aircraft delivered, more than 1 million flight hours, 51 bases worldwide, more than 3,290 pilots trained, and more than 20,100 maintenance personnel trained. At this scale, every software improvement counts enormously. A useful new AI feature doesn’t just impact one squadron. It can ultimately transform a global fleet. This is why the F-35 remains central to the United States and its allies: the software scale effect itself becomes a strategic advantage.
What allies will actually receive in terms of AI
In principle, the official response is quite clear: the Block 4 upgrades apply to all three variants and benefit all nations operating the F-35. Lockheed has repeated this several times. If an AI capability is integrated as part of a validated Block 4 increment, the logic of the program therefore pushes for widespread distribution within the F-35 ecosystem, and not just for isolated US use.
But the reality is more subtle. What will be “sold” to allies will not necessarily be a uniform package of freely modifiable AI. The F-35 remains a highly centralized system, driven by a common architecture, software standards, update cycles, and classified components. Allies benefit from the common base, yes. However, the depth of access, the pace of integration, sensitive data, certain threat libraries, and some of the most sensitive functions remain restricted. This is the whole issue with mission data files.
The case of reprogramming laboratories is revealing. In 2024, the United Kingdom, Australia, and Canada strengthened their partnership within their joint mission data center at Eglin. The F-35 website points out that these mission files enable F-35s to detect, identify, locate, and counter threats across the electromagnetic spectrum. In other words, AI is worthless without good reference data. And this data is subject to a chain of partial sovereignty, alliance, and validation.
Allies will therefore receive more advanced capabilities, but within a highly structured framework, unlike civilian software that can be modified at will.
Two opposing errors must therefore be avoided. The first is to believe that allies will receive an “empty” or degraded version of any software advances. This is false: the value of the F-35 program is based precisely on standardization, interoperability, and the critical mass of the common fleet. The second is to believe that each customer will have total sovereignty over all AI building blocks, models, and data flows. This is equally false. The F-35 is a coalition platform, but also an organized dependency architecture.
What the term “AI” still hides in the public debate
Industrial discourse naturally tends to broaden the scope of the term “AI.” However, with the F-35, the decisive point is not whether an algorithm bears this label. The decisive point is whether it actually reduces the time between detection, understanding, and decision-making. If the answer is yes, the capability has military significance. If the answer is no, it is just another sales pitch.
The F-35 is entering a new phase. The historical building blocks of data fusion and decision support are already in place. AI building blocks are beginning to emerge from the laboratory and enter serious tactical demonstrations. The trajectory is clear: better identification, better electronic warfare, better coordination with unmanned systems, and faster reprogramming cycles. But the road ahead remains constrained by a brutal reality: the F-35’s software modernization is cumbersome, costly, and behind schedule. This will continue to prevent any rapid shift to “ubiquitous AI” on board for some time to come.
What is at stake here is broader than the F-35. If the program succeeds in industrializing these new capabilities across a fleet of more than 1,300 aircraft, it will set a standard for the next decade: that of a fighter that dominates not only through its stealth or missiles, but through its ability to absorb, interpret, and redistribute information faster than any other. And in modern air combat, that is probably the true definition of sustainable air superiority.
Sources
Lockheed Martin / F35.com, Lockheed Martin Applying AI to Enhance F-35 Combat Identification System, February 23, 2026.
Lockheed Martin / F35.com, Block 4 Capabilities Sharpen the F-35’s Edge, August 1, 2024.
F-35 Joint Program Office / Air Force Materiel Command, F-35 Conducts First Flight with TR-3, January 10, 2023.
U.S. Government Accountability Office, F-35 Joint Strike Fighter: Actions Needed to Address Late Deliveries and Improve Future Development, September 3, 2025.
U.S. Air Force, F-35A Lightning II Fact Sheet.
Lockheed Martin / F35.com, 5th Gen Capabilities.
Lockheed Martin / F35.com, Fast Facts, updated 2026.
Lockheed Martin / F35.com, Allies Strengthen F-35 Mission Data Partnership, April 24, 2024.
Lockheed Martin, F-35A Product Card.
BAE Systems, AN/ASQ-239 F-35 EW countermeasure system.
War Wings Daily is an independant magazine.