A fighter aircraft flown by software is not just an aviation story. It changes where value sits inside the aircraft.
The US Defense Advanced Research Projects Agency and the US Air Force have tested an AI-controlled F-16 under the VENOM programme, with a safety pilot able to take control. Capital described the flight as another step toward coordination between human crews and autonomous aircraft.
The important detail is the supervision model. Autonomy in a high-risk aircraft is not a switch between human and machine. It is a chain of permissions, limits, observations and interventions. The aircraft may control a manoeuvre while a human retains authority over the mission. Engineers need to prove what happens when sensors disagree, communications fail or the software encounters a situation outside its training.
France already understands collaborative aerospace
France is not starting from zero. Dassault Aviation led the nEUROn unmanned combat-air demonstrator with European industrial partners. Thales works across sensors, secure communications and decision-support systems. Safran supplies flight, navigation and propulsion technologies. Research institutions and defence procurement agencies add another layer of expertise.
Yet a list of capable organisations is not an autonomous system. The difficult work sits between them: common test environments, interfaces that can change without grounding the fleet, access to representative data and a certification process that can cope with software behaviour.
Traditional aircraft programmes are organised around long hardware cycles. AI systems change through data, evaluation and updates. That creates tension. A military customer wants a stable, auditable configuration. A learning programme wants frequent iteration. France’s aerospace model will have to support both without handing operational control to an opaque software stack.
Sovereignty moves into the training loop
European arguments about defence sovereignty often focus on where equipment is assembled. For autonomous systems, sovereignty also means knowing which data trained a model, who can update it and whether performance can be measured independently.
A country can own the airframe and still depend on someone else for the decision layer. Conversely, it can build excellent software that remains unusable because it cannot be integrated safely with sensors, weapons controls and command networks. The competitive unit is the whole verified system.
This favours organisations that invest in simulation and instrumented testing, not only in impressive demonstrations. Millions of virtual encounters can expose rare edge cases, but simulation itself must be checked against flight data. The feedback loop between virtual and real tests becomes an industrial asset.
The export question is trust
French combat aircraft have an international market because customers value performance, political alignment and a degree of operational independence. Autonomous capabilities add a harder sales question: will the buyer understand and control the system well enough to trust it?
Exportable autonomy will need clear operating limits, local training and evidence that updates do not silently change behaviour. Customers may demand sovereign data environments or the ability to validate software themselves. That is expensive, but it can become a French and European differentiator if larger competitors offer more closed systems.
The American test does not mean pilots are about to disappear. It means software is becoming a crew member, and every aerospace power now has to decide how that crew member is trained, supervised and trusted.
