Aviation AI's Certification Problem Isn't the Algorithm

The question regulators are actually stuck on

The industry conversation about AI in aviation keeps drifting toward capability. Can the model fly the approach, taxi the aircraft, manage the airspace conflict. That is the wrong axis. The harder question, and the one insurers and regulators are visibly still working through, is who is accountable when the system acts. Global Aerospace’s brief on how regulators are approaching autonomous flight systems makes clear that basic questions about liability and insurance attachment for autonomous decisions remain open even as the underlying systems mature.

That gap shows up in how companies are routing around it commercially. Merlin abandoned its New Zealand autonomy program to instead pursue roughly 7,000 US airline cockpits, shifting from full autonomy certification toward assisted operation inside existing, already-certified crew structures. That is not a retreat from the technology. It is a bet that the accountability chain, not the algorithm, is the faster path to revenue.

Where the frameworks actually point

This is a governance problem with named frameworks attached, not an abstract one. ISO 42001 style AI management systems, EU AI Act human oversight obligations for high-risk systems, and NIST AI RMF’s govern and manage functions converge on the same requirement: a documented, auditable record of who decided, on what basis, and who retains override authority. Aviation certification has no settled equivalent for AI-in-the-loop decisions yet, which is the vacuum Global Aerospace is describing from the insurance side.

Defense programs are testing the same problem under different constraints. Dassault Aviation is running AI algorithms on the Rafale with a human pilot still in the aircraft, and at TechCrunch Disrupt, Shield AI, Waabi, and General Motors discussed building AI systems for domains where failure is not an option, treating accountability architecture as a design constraint rather than a compliance afterthought. As one defense commentary put it, autonomy does not remove the need for a signature, it relocates it. Aviation’s certification bodies have not yet decided where that signature lives.

The broader AI assurance environment is not making this easier to wait on. A UN panel has warned that current AI guardrails are unraveling following the OpenAI cyberattack, a reminder that the assurance infrastructure aviation regulators are drawing on is itself under strain elsewhere in the AI stack.

The decision in front of leadership

For program leads and regulatory affairs teams, the practical move is to build the accountability architecture, the audit trail, the defined override authority, the documented sign-off chain, before an aircraft needs it, not after a regulator asks for it. Map ISO 42001 documentation and EU AI Act oversight obligations against your actual flight-deck decision points now, while certification frameworks are still being written rather than enforced.

The companies that treat this as engineering work, not paperwork, will be the ones cleared to fly first.

---

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 core argument—that accountability architecture, not algorithmic capability, is the binding constraint on aviation AI certification—is coherent and well-supported by the Merlin pivot and Global Aer
Source & Claim Verification	Qwen · hosted fallback (qwen-2.5-72b-instruct)	cleared. Most factual claims are supported by citations, but a few lines could benefit from more specific references or additional context.
Regulatory & Framework Fidelity	Mistral	cleared. The brief accurately reflects the accountability and governance requirements of ISO 42001, EU AI Act, and NIST AI RMF, but does not substantively address 21 CFR Part 11 or EU GMP Annex 11/22, which ar
Technical Accuracy	Llama	cleared. The article accurately captures the current challenges in certifying AI in aviation, particularly around accountability and liability, and correctly references relevant standards and regulatory framew
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively counters potential vendor hype by focusing on the regulatory and accountability challenges rather than technological capabilities, and it uses external sources to support its
Novelty & Non-Duplication	Grok	held. The accountability-over-capability thesis is a stock AI-governance frame; this mostly repackages same-week wire (Merlin pivot, Global Aerospace insurance brief, Dassault/Rafale, Disrupt panel) without
Validation	DeepSeek	cleared. The central claim that aviation AI certification’s core blocker is accountability/liability frameworks, not algorithm capability, is strongly validated by industry actions, regulator/insurer statement

Sources cited: 11. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.

Record corrected 29 September 2026. This seat was originally listed as running on our own local hardware. It never did: every review before this date fell back to a hosted model, and the row above now names the model that actually ran.