The Pentagon Is Putting Agentic AI Aboard the Next Doomsday Plane

The U.S. Air Force is putting agentic software on the jet built to keep nuclear command alive if the ground goes dark. On October 2, 2026, the Department of the Air Force awarded Scale AI Inc. a $12,093,115 firm-fixed-price modification (P00003) to contract FA875126C0001 for Agentic Artificial Intelligence on the E-4C Survivable Airborne Operations Center program.

The modification lifts the cumulative face value of Scale AI's SAOC contract from $32,247,194 to $44,340,309. Air Force Life Cycle Management Center at Hanscom Air Force Base, Massachusetts, is the contracting activity. At award, the service obligated $5,000,000 in fiscal 2026 research, development, test, and evaluation funds. Work is set for San Francisco, California, through November 5, 2027.

Twelve million dollars will not buy a doomsday plane. It can buy the software layer that decides whether the next one is a flying conference room or a command post that still functions after satellites fail.

Nightwatch was built for a different war

The E-4C is the replacement for the Air Force's four E-4B Nightwatch aircraft, popularly called doomsday planes, which have flown since the 1970s on Boeing 747-200 airframes. Those jets exist so the President, the Secretary of Defense, and the Joint Chiefs of Staff can still run nuclear command, control, and communications if fixed sites are destroyed.

In April 2024, the Air Force awarded Sierra Nevada Corporation a $13.08 billion contract to develop the Survivable Airborne Operations Center, beating incumbent Boeing. SNC acquired five commercial Boeing 747-8 passenger jets from Korean Air to convert into hardened military command posts.

That industrial bet is conversion: take a commercial widebody, armor the mission systems, and make it a national airborne operations center. The software bet is different. Battle staff on that aircraft will have to sort intelligence, keep communications moving, and present options while an adversary is trying to blind every link the staff still has.

The E-4B grew up in an analog, isolated computing culture. Radios, teletypes, and tightly controlled circuits were features, not bugs, because the mission was continuity of government under nuclear attack. The E-4C is being asked to do that job in a world of multi-domain sensors, contested electromagnetic spectrum, and data volumes no 1970s crew could have processed by hand.

Agentic tools, open architecture, human authority

The SAOC program was structured with Modular Open Systems Architecture so the jet would not be locked to a single software vendor for decades. Scale AI's agentic tools are designed to process large, multi-domain intelligence feeds, route critical communications under contested or degraded electromagnetic conditions, and help military battle staff make faster operational decisions when satellite and ground links are severed.

Agentic systems can draft courses of action, prioritize traffic, and stitch sensors together. Nuclear command still requires a human in the loop. The engineering problem is speed without handing launch authority to a model. The political problem is the same one that has always sat inside NC3: the President and senior commanders must be able to see the picture, issue orders, and be heard, even if Washington is gone.

That tension is the real content of the Scale AI modification. A firm-fixed-price increment through November 2027 is a software clock running inside a much larger recapitalization. Sierra Nevada has to deliver hardened 747-8s. The Air Force has to field an airborne node that still works on the first day of a major war. Scale AI has to show that agentic models can sit in that node as staff aids, not as a single point of failure.

Open architecture is supposed to make that possible. If the government owns the interfaces, it can swap models, vendors, and data pipelines as the threat changes. If the interfaces are theater, the E-4C inherits a new kind of lock-in: not to a 747-200 airframe this time, but to whatever software stack gets wired in first.

Survivability is now a software problem too

Strategic survivability used to mean hardening, tankers, and radios that could punch through nuclear effects. Those still matter. The next gap is cognitive. A flying command post that cannot synthesize what its own sensors and allied feeds are saying is not survivable in a practical sense, even if the hull stays aloft.

The Air Force is therefore doing two recapitalizations at once. One is metal: five 747-8s converted into command posts under a $13.08 billion Sierra Nevada contract. The other is code: agentic software on a $44.3 million Scale AI vehicle, now expanded, with a San Francisco performance line that runs into late 2027.

Allied nuclear command has always treated automation with suspicion, and it should. Agentic tools on the E-4C exist because human decision-makers cannot be the bottleneck for routing, correlation, and staff work when the spectrum is hostile. If Scale AI's work holds to that line, the next doomsday plane will still be a human command post, one that no longer pretends 1970s computing is how you fight a 2020s war.