Aviation Is Building a Second Certification Track for Autonomy

Aviation Is Building a Second Certification Track for Autonomy

Shield AI’s V-BAT just earned Specific Assurance and Integrity Level III, the highest SAIL rating granted for maritime unmanned operations in Europe, during a Frontex maritime surveillance mission startuphub.ai. That is not an airframe type certificate. It is a risk-tiered assurance rating for the operation itself, covering autonomy behavior, mission software, and the ways the system can fail, layered on top of whatever airworthiness basis the aircraft already holds. For regulated buyers, that distinction is the story.

Traditional aerospace certification asks whether the airframe and its systems meet a design standard. SAIL, as EASA applies it to unmanned and increasingly autonomous operations, asks a different question: given what this system does on its own, how much can go wrong, and what mitigations exist. That is a governance question, not just an engineering one, and it sits closer to ISO 42001’s risk-based AI management logic than to classical airworthiness review.

The infrastructure choices behind that autonomy are shifting in parallel. NVIDIA’s new TensorRT Model Connect compiles models straight from a Hugging Face checkpoint into native C++ inference, explicitly targeting defense and aerospace edge systems where inference has to run inside a compiled binary rather than a Python server marktechpost.com. That is not a performance detail. A Python runtime with a sprawling dependency tree is close to unauditable inside a safety case. A compiled, deterministic binary is something an assurance reviewer can actually inspect. The industry is quietly rebuilding its inference stack to be certifiable, ahead of any regulator asking it to.

The advisory layer is moving the same direction. Google and NATS are running a North Atlantic trial where machine learning models forecast contrail formation to guide route changes for warming avoidance, with NATS retaining responsibility for the actual air traffic decisions aerospacetestinginternational.com. The AI model never touches the aircraft. It informs a human-controlled system. That is a third assurance category again, distinct from both airframe cert and SAIL-style autonomy rating, and current frameworks don’t clearly say which one governs an advisory model feeding safety-relevant routing decisions.

The decision for airlines, defense primes, and air navigation providers buying into this stack is no longer just “is the aircraft certified.” It is a two- or three-track question: what is the airframe’s airworthiness basis, what SAIL or equivalent rating covers the autonomy and inference stack running on it, and what governance regime covers any AI system advising a human operator rather than acting directly. Vendors who can only answer the first question are selling last decade’s product. The V-BAT rating, the compiled inference toolchain, and the contrail trial are three separate signals of the same shift: assurance is decoupling from the airframe, and buyers who keep asking a single certification question will miss the two new ones.

---

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 aviation assurance is fragmenting into multiple certification tracks beyond traditional airworthiness—is coherent and well-supported by three distinct examples, though the cl
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but the article could benefit from more explicit references to the sources within the text.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects ISO 42001’s risk-based AI management logic and EASA’s SAIL framework but does not explicitly address FDA, MDR/IVDR, or EU AI Act conformity requirements for aviation A
Technical Accuracy	Llama	cleared. The article accurately describes the shift in aviation certification towards a multi-track approach for autonomy, with a clear understanding of SAIL ratings, airworthiness, and the implications of AI
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and critiques vendor hype by distinguishing between traditional certification and new, risk-tiered assurance models, and by highlighting the practical implications
Novelty & Non-Duplication	Grok	cleared. The multi-track assurance thesis is a genuine synthesis of three separate wire items rather than a rewrite of any one, and nothing in the offered source set or the framing itself signals catalogue dup
Validation	DeepSeek	cleared. The briefing demonstrates a decoupling of autonomy governance from airframe certification through concrete examples, though it overstates the maturity and pervasiveness of a ‘second track’.

Sources cited: 12. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.