When the Coder Is an Unrated Component

When the Coder Is an Unrated Component

Aerospace software has never operated on trust. DO-178C exists precisely because a compiler bug in a flight control system is not a bug report, it is a hazard analysis. Every line of code that touches flight-critical behavior gets traced, verified, and signed off before it flies. That discipline is now colliding with a new category of contributor: the AI coding agent that writes, modifies, or deploys software with no human in the loop at the moment of change.

A recent framing from IT Business Net makes the stakes explicit. Once an AI system can modify or deploy code that affects operational behavior, that is no longer a developer convenience. It is an instance of autonomy that must satisfy the same verification, validation, and safety criteria as any other autonomous system operating in a regulated environment. The piece calls for treating software itself as a governed, dynamic asset rather than a static artifact that gets certified once and left alone.

For aerospace program offices, this is not a hypothetical. Airlines and suppliers are already piloting AI-assisted code generation and maintenance tooling inside environments subject to FAA and EASA software assurance regimes. The question those regulators will eventually ask is not whether the AI is good at writing code. It is whether the code it writes, and every subsequent change it makes, can produce the same traceable verification record that DO-178C already demands of human engineers. An AI agent that silently patches a build without generating that record is, in certification terms, an unrated component sitting inside a rated system.

The liability question sharpens the point. A recent commentary in Automotive News argues that autonomous vehicle liability frameworks should borrow directly from aviation’s model of shared accountability across manufacturer, operator, and regulator. The same logic applies to autonomous software agents. If an AI agent modifies flight-adjacent code and that change contributes to a failure, the question of who bears responsibility, the model provider, the integrator, or the airline’s engineering sign-off, has no settled answer yet. Aviation built decades of case law and certification precedent to answer that question for human engineers. It has none for agents.

ISO 42001 gives enterprise AI governance teams a starting management-system structure, but it was not written with DO-178C’s line-by-line traceability in mind. The EU AI Act’s high-risk classification will likely sweep in AI systems embedded in safety components, which puts flight software squarely in scope, but implementation detail on code-generation tooling remains thin.

The decision in front of regulated buyers is narrow and specific. Before any AI coding agent gets write access to production systems in a certified environment, someone needs to define what verification record that agent must produce, and who signs it. Aviation spent seventy years building that answer for humans. It has not yet built it for machines that write the code humans used to write.

---

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 AI coding agents in certified aerospace environments create an unresolved verification and liability gap—is coherent and logically constructed, but the piece overstates regulato
Source & Claim Verification	Qwen · local	cleared. All factual claims are traced to citations, but some sources (e.g., China’s C919 flight articles) are not directly relevant to the main argument and could be more focused.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects the regulatory tensions between AI coding agents and ISO 42001/EU AI Act requirements but lacks explicit mapping to FDA/MDR/IVDR contexts where applicable.
Technical Accuracy	Llama	cleared. The article accurately captures the challenge of integrating AI coding agents into aerospace software development under DO-178C and highlights the need for verification and validation processes for AI
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies a core concern regarding AI coding agents in regulated environments and largely avoids vendor hype, though some sources lean promotional.
Novelty & Non-Duplication	Grok	held. Core thesis is largely a restatement of the cited IT Business Net “governed autonomy” framing plus the Automotive News liability parallel, with standard DO-178C gloss; little original reporting or ang
Validation	DeepSeek	cleared. The central claim that AI-generated code in certified systems currently lacks a defined verification and liability framework is validated by the cited sources on governed autonomy and liability models

Sources cited: 13. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.