The U.S. Air Force Collaborative Combat Aircraft effort is real and moving into an operational test window that matters for anyone who flies fighters. General Atomics’ YFQ-42 production-representative vehicle has entered ground testing with first flights expected later in the summer, and the Air Force has already set up an Experimental Operations Unit to exercise human-machine teaming concepts. These are the conditions under which real pilots and real autonomous wingmen will have to operate together, and that reality demands a straightforward, operational safety checklist before any manned-unmanned teaming sortie becomes routine.
Start from the basics: command, control, and authority. Human pilots must retain clear, immediate authority over offensive and defensive actions that could escalate a fight. In practice this means cockpit procedures that specify the handover sequence, positive confirmation of mission state, and unambiguous abort and safe-recovery commands. Test plans should require simple, robust command verbs and hardwired timeout behaviors so the platform defaults to a conservative, non-threatening state if the datalink goes flaky in a contested environment. Those behaviors need to be exercised in worst case communications scenarios during flight test, not just in simulation.
Sense and deconfliction are the next priorities. A CCA operating with an F-22 or F-35 must share tracks and intent without creating new collision hazards. That requires validated, latency-bounded track-sharing and a prioritized deconfliction logic that mirrors what pilots already trust in TCAS and formation flying. For example a wingman AI should yield to an indicated collision avoidance command from the crewed fighter and must never autonomously execute a high risk maneuver in a busy formation without explicit pilot authorization. Test sorties should include formation rendezvous and breakaway drills, with aircrew practicing last-second overrides while controllers monitor separation margins in real time.
Autonomy needs boundaries and explainability. Black-box machine learning policies can work well in constrained environments but become dangerous if they are allowed free rein in air combat. Flight test protocols should force the autonomy to declare its confidence envelope and fall back to pre-approved tactics when confidence is low. Recording and post-flight forensics will be essential; every autonomous decision that affects flight path or weapons employment needs logged telemetry, sensor inputs, and the AI rationale to support after-action review and any regulatory or legal inquiries. The Experimental Operations Unit model is the right idea because it embeds operational pilots and testers together to iterate those requirements.
Communications resilience is non-negotiable. The CCA concept depends on timely, secure datalinks. Tests must include degraded, jammed, and spoofed scenarios to validate fail-safe modes. Practical mitigations I look for in test programs are: multi-path comms (line-of-sight and beyond-line-of-sight redundancies), prioritized safety messages that use minimal bandwidth, and a preplanned behavioral palette for comms-denied conditions (return-to-base, loiter and proceed, or execute safe loiter over defined airspace). Those behaviors should be simple, predictable, and identical across vendor platforms so a pilot transferring between teeming aircraft does not have to relearn safety logic.
Human factors and cockpit integration matter. The pilot workload in an F-35 or F-22 is already high in combat and will increase when you have unmanned teammates to manage. UI design must avoid modal traps and should present only the essential status items: teammate health, intent/mission state, and immediate safety warnings. Any manual override must be single-action and hard to mis-execute under G or stress. Flight test should include human-in-the-loop evaluations under realistic mission pressure so that interface changes are driven by pilot performance data rather than laboratory ergonomics.
Airworthiness and maintenance safety are often overlooked in the rush to field capability. The CCA is intended to be production-representative which is a double-edged sword: you get realistic systems earlier, but you also need to validate the sustainment chain in the same window. Safe fielding requires that the maintainers have standardized fault trees, test jigs, and simple safe recovery procedures for in-flight anomalies. Safety during flight test is only as strong as the ground-handling and preflight checks that precede it.
Rules of engagement and legal compliance must be baked into the autonomy stack. That means weapons release logic must be auditable, constrained to commander-authorized states, and subject to immediate human veto. For tests that do not involve live weapons, simulated weapons employment must follow the identical decision logic and logging rules as a live employment would, so testers see any edge-case behavior before a lethal decision is possible. The Experimental Operations Unit provides a venue to marry operational doctrine with technical capability during the test phase.
Finally, treat flight testing as a learning campaign not a demonstration. The CCA program has been set up with incremental prototyping, vendor testing, independent evaluation, and operational assessments. That same phased approach should be the safety model: throttle complexity, validate simple building blocks, and only then combine them into contested, high-speed engagements. Introduce formation flying, low-altitude maneuvering, and weapons integration on a laddered timeline with explicit go/no-go criteria at each step and independent safety oversight.
If you are a pilot who will fly with the YFQ-42, or a test planner responsible for the sorties that put it in the air, push for these practical measures: clear authority transfer procedures, mandatory logging of autonomy decisions, real-world degraded-communications testing, pilot-centric UI design, maintainers’ readiness, and stepwise escalation of mission complexity. Treat AI teaming as an operational system that must be proven safe in the air, on the ground, and in the cockpit. Do that and the promise of affordable, effective collaborative combat aircraft becomes deliverable in a way that protects pilots and preserves the trust required for manned-unmanned teaming to work in combat.