Inside DARPA's Air Combat Evolution program, which aims to design AI that can fly a plane and engage in aerial combat without a human pilot
Sue Halpern / New Yorker : Tweets: @brandonhamber , @michaelluo , @newyorker , and @hoanssolo Tweets: Brandon Hamber / @brandonhamber : While people fight over masks and freedoms, artificial intelligence is already driving a worldwide arms race and “allowing machines to choose to kill humans will be devastating to our security and freedom” https://www.newyorker.com/... https://twitter.com/... Michael Luo / @michaelluo : Top Gun. AI-style. @suehalpernVT on devising an algorithm to dogfight. https://www.newyorker.com/... @newyorker : Artificial intelligence is being taught to fly war planes. How much autonomy do we want for a system that can deliver lethal force at machine speed? https://nyer.cm/IuCHNxD Franz-Stefan Gady / @hoanssolo : There are currently more than 600 Pentagon projects that are trying to incorporate artificial intelligence into war-fighting. https://www.newyorker.com/...
Context & Ripple Effects
Sue Halpern's New Yorker piece lands mid-arc for DARPA's Air Combat Evolution program: the ethical questions about how much kill authority machines should get were already being debated in Washington when the program was teaching algorithms to dogfight, and the international standoff over a ban — 30 countries in favor, the US dismissive framed the stakes. What makes the profile matter in hindsight is what came after it.
Within roughly a year of publication, DARPA reported that ACE's AI had flown a modified F-16 test aircraft multiple times, and by 2023 an AI-controlled X-62A had faced a human pilot in an actual in-air dogfight. The article captured the program at the moment it moved from simulation to hardware — the pivot point where the ethics debate stopped being hypothetical.
First-order effects
- DARPA gains proof that its autonomy stack can handle real flight dynamics, converting ACE from a research bet into a candidate for operational programs — with the X-62A/F-16 test fleet as the demonstration vehicle.
- Human pilots' role in future air combat becomes an open procurement question rather than an assumption, since the program's explicit goal is aerial combat without a human in the cockpit.
Second-order effects
- Nontraditional suppliers gain leverage: coverage of Shield AI's funding hurdles against entrenched weapons makers shows the Pentagon's autonomy demand pulling startup-grade software into a prime-contractor market built around platforms.
- The US position that autonomous-weapons concerns are 'overblown' hardens into policy as capability matures, putting it on a collision course with the bloc of countries pushing a total ban.
Third-order effects
- If lethal decisions migrate to machine speed, the unresolved question from the 2021 ethics debate — how much control humans retain over the kill chain — becomes the central regulatory fight, not a design footnote.
- Air power structurally shifts toward attritable, software-defined aircraft where the differentiator is the model, not the airframe — favoring defense primes that can absorb AI firms or partner with them.
The trend: Military aviation is moving from human-piloted platforms toward AI-flown ones faster than the international governance debate over autonomous lethal force can keep up.