Air Combat Command's Experimental Operations Unit (EOU) concluded a critical exercise this spring at Edwards Air Force Base in which Air Force airmen — not General Atomics or Anduril test crews — took direct control of Anduril's YFQ-44A Fury through pre- and post-flight checks, weapons loading and unloading, taxi, and tasking during flight. The 412th Test Wing under Air Force Materiel Command provided range and instrumentation support, but the operating tempo was set by frontline operators working out of what the service described as a simulated forward operating base. On June 9, Secretary of the Air Force Troy Meink told the Senate Appropriations Committee the CCA program will expand its supplier base, sustain its open architecture posture, and pursue scale procurement in Increment 2. The two events should be read together. They are the same bet, stated in two different forums.
The bet is that the Air Force can replace the traditional acquisition sequence — requirements, source selection, development, operational test, fielding — with a tighter loop in which operators, acquirers, and vendors share the same airframes during development. The service has given this loop a name: the Warfighting Acquisition System. CCA is its first major pathfinder. Anduril's Menace-T command-and-control ground element ran the EOU exercise without a fixed infrastructure footprint, aligning with Agile Combat Employment doctrine and letting the unit shift between sortie profiles in days rather than program reviews. That is a deliberate signal to industry. Anything in the CCA program that requires a contractor-staffed mission control center, a proprietary ground station, or a vendor-only weapons handling crew will lose against systems that can be deployed and sustained by line airmen with standard tooling.
What the Open Architecture Push Actually Buys
Secretary Meink's June 9 testimony framed open architecture as a cost play — more suppliers, more competition, lower unit cost. That framing undersells what the Air Force is actually doing. Open architecture in the CCA context is the condition that lets the EOU loop run at all. If the autonomy stack, mission management software, and weapons integration interfaces are vendor-locked, then every operator-discovered tactic, every new payload, and every contested-environment lesson learned at Edwards has to be routed through the prime's release schedule. That kills the entire premise of operator-driven experimentation. The Air Force needs the right to modify, recompile, and re-host CCA software on its own cadence, with multiple integrators contributing to the same reference architecture. That is a much harder ask than competitive bidding on airframes.
The Block 3 MQ-28 Ghost Bat that Boeing revealed at ILA Berlin this week — larger wing, internal weapons bays, AIM-120 carriage — is a reminder that the airframe side of CCA is moving fast and crowded. General Atomics' YFQ-42A returned to flight test in May after a software-related mishap, with autopilot weight-and-balance handling fixed. Anduril's YFQ-44A is in serial production at Arsenal-1 and now in operator hands. The hardware competition is real but increasingly commoditized. The decisive contest is over whose autonomy stack, whose mission planning runtime, and whose verification regime the Air Force certifies to ride on top of multiple airframes from multiple primes — and whose can be modified by the government and its system integrators without breaking a support contract.
The Verification Problem the EOU Loop Just Exposed
Operator-driven experimentation accelerates fielding only if the autonomy underneath behaves predictably enough that an EOU airman can task it during a sortie without a contractor-supplied flight test engineer in the loop. That is a far higher bar than what most autonomy stacks have been built to meet. Today's CCA autonomy is largely opaque: the operator gives high-level intent, the system executes, and post-mission analysis tells the test team whether the behavior was what they expected. When the loop closes inside a single flight, with operators making real tasking decisions on a contested-environment simulation, the verification gap shows up immediately. Behaviors that pass scripted test cards do not necessarily pass operator improvisation. Models that are accurate on training distributions do not necessarily generalize to the contested-environment overlays the EOU is exercising. The April 6 YFQ-42A mishap, traced to an autopilot weight-and-balance miscalculation, was a clean engineering failure with a clean software fix. The harder failures — the ones that will define whether the Warfighting Acquisition System can deliver on its premise — will be autonomy decisions that look reasonable in isolation but fall outside the doctrine an operator is trying to apply.
This is the layer Spartan X has built Arbiter to address. Multi-model consensus, behavioral verification against a separate adversarial model, and a verification record that travels with the autonomy stack from contractor flight test through EOU operations and into operational squadrons. The CCA program is making the policy choice that autonomy verification will be a government-curated, government-modifiable property of the reference architecture rather than a deliverable that ships once with a prime's release. That choice is consistent with how the broader DoD AI verification posture is evolving — the NDAA AI sandbox provisions, the CSRMC end of static ATO, the agentic-AI verification gaps surfaced in the DoD's gen-AI evaluation work — and it is the right one. The CCA program is now the place where that policy gets tested under operational tempo. The vendors and integrators who treat verification as a feature of the autonomy stack — not a checkpoint at delivery — are the ones who will still be in the CCA reference architecture three increments from now.
