Where Aerospace AI Is Already Flying, and Where It Still Can't

Two different airplanes

Aerospace AI is not one story right now. It is at least two. At Incheon International Airport, an AI system is already checking aircraft safety and giving directions to ground crews and travelers, operating today, in production, under ground-operations oversight rather than flight-critical certification. Meanwhile, the software that actually flies the aircraft is still bound by DO-178C and DO-330, the standards written before anyone was trying to certify a statistical model. Boards tracking a projected $57.6 billion shift in aerospace AI spending need to know which side of that line their investment sits on, because the assurance burden is completely different depending on the answer.

The gap that doesn’t close on its own

For anything on the flight-critical side, the problem is not paperwork, it is physics of a different kind. AI models behave statistically, and that behavior has to be characterized before anyone can build a credible case around it, a gap researchers have flagged directly: certification methods built for deterministic software were not built for this. That characterization work is now specialized enough to warrant its own research track, evident in dedicated academic attention to the subject from institutions like the Aerospace Corporation. This is no longer a side note to the certification process. It is becoming a discipline of its own, running in parallel to it.

Regulatory throughput has not caught up to either track. ASD Europe’s figures put EASA certification staffing at roughly 280 full-time equivalents against the FAA’s 1,400, a gap closer to fivefold than four. That constrains how much novel, judgment-intensive review either body can absorb, full stop. It says nothing about ground-ops deployments like Incheon’s, which don’t sit in that queue at all, and that asymmetry is itself a decision point: capital and deployment are moving fastest exactly where certification friction is lowest.

What that means for capital allocation

Investor interest in the sector has not waited for the standards to catch up either, with defense and aerospace names continuing to draw hedge fund attention according to recent portfolio data. Merlin’s framing of assured autonomy is a vendor’s pitch, but the underlying claim tracks: operators buy demonstrated capability legible to an existing certification authority, not a bet on a future rulebook. SESAR’s MEDUSA project argues the same thing from the standards side, that aviation AI needs continuous assurance rather than one successful validation run.

The decision

Know which side of the certification line each deployment sits on before building its evidence package. Ground-ops AI like Incheon’s can move at commercial speed and should be evaluated on operational metrics, not DO-178C. Flight-critical AI cannot, and its evidence has to be built against DO-178C and DO-330 as they stand, with the statistical characterization work funded as its own line item rather than folded into general R&D. Regulatory staffing will not speed up. The boundary between the two tracks, and the discipline to fund each correctly, is what a program actually controls now.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The central argument—that aerospace AI bifurcates into ground-ops (fast, commercial metrics) versus flight-critical (slow, DO-178C bound)—is coherent and well-supported, though the hedge fund portfoli
Source & Claim Verification	glm-5.3-flash-fp4 · local	cleared. All quantitative anchors verify against their opened sources (Incheon’s AI safety-check/directions system per the MK headline, the $57.6B-by-2033 figure per Futurium, and EASA’s 280 vs FAA’s 1,400 FTE
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately distinguishes between ground-ops AI (not subject to DO-178C) and flight-critical AI (requiring DO-178C/DO-330 compliance), aligning with aerospace regulatory frameworks.
Technical Accuracy	Llama	cleared. The article accurately distinguishes between the certification requirements for ground-operations AI and flight-critical AI in aerospace, highlighting the need for statistical characterization of AI m
Bias, Balance & Hype Control	Gemini	cleared. The briefing generally avoids vendor hype, but the inclusion of Merlin’s pitch, even with a disclaimer, introduces a slight bias.
Novelty & Non-Duplication	Grok	held. Spine restates the 2026-10-31 EASA staffing-bottleneck and DO-178C/flight-AI gap briefings; Incheon is new color but not a genuinely new fact or decision beyond that catalogue.
Validation	DeepSeek	cleared. The central claim of a clear functional and regulatory divide between ground operations AI and flight-critical AI is strongly validated by current, disparate certification practices and regulatory sta

Sources cited: 9. Validation challenges: 0. Learn how these briefings are written and verified.