Washington's AI Pause Exposes a Gap DO-178C Was Never Built to Close

More than two dozen lawmakers have asked the FAA to pause a new AI tool deployed to assist air traffic controllers in the Washington-area airspace, citing safety concerns tied to the agency’s rushed pilot rollout and an existing controller shortage The Jerusalem Post. House Democrats have gone further, pushing for a Government Accountability Office audit of the FAA’s AI-enabled air traffic software, with the request landing barely a month after the pilot began Federal News Network.

The specific tool matters less than what the pause request exposes: there is no certification basis today that actually covers what this software does. DO-178C has long served as the certification basis for airborne software, built on requirements-based testing and structural coverage against deterministic, traceable logic paths. That model assumes you can enumerate the decision space in advance and verify every path through it. Learned, probabilistic decision support does not fit that assumption cleanly, and nothing has replaced DO-178C’s role for AI components in safety-of-life systems. Lawmakers are not objecting to AI in principle. They are objecting to the absence of anything equivalent to certify against.

The frameworks in play don’t close the gap alone

ISO 42001 gives organizations an AI management system, a structure for policies, risk treatment, and continual improvement. It is a governance standard, not a flight-worthiness certification basis. NIST AI RMF’s govern-map-measure-manage functions supply a vocabulary for documenting training data lineage, decision logging, and human override, but the framework is voluntary and was not written with air traffic control in mind. The EU AI Act’s high-risk categorization for safety components of regulated transport systems will eventually pull aviation AI into a conformity assessment regime, but that regime was built for product safety directives, not for the kind of airworthiness certification DO-178C performs. Each framework covers part of the problem. None of them, alone, gives FAA or EASA something they can certify a learned system against the way they certify conventional avionics software today.

This reframes what FAA-EASA cooperation on AI roadmaps actually needs to produce. The two agencies have published roadmaps and industry observers see convergence toward a shared technical understanding of emerging risk Commercial UAV News. The harder test is whether that cooperation yields a certification basis that absorbs ISO 42001-style governance evidence and NIST AI RMF-style measurement artifacts into a form DO-178C-trained certification engineers can actually audit. That synthesis does not exist yet, which is precisely why a congressional pause request could land with no finalized standard to answer it.

The commercial AI assurance market is already building adjacent infrastructure. The AICPA and the Artificial Intelligence Underwriting Company have announced a combined SOC 2 and AIUC-1 certification aimed at giving enterprises independent review of AI-related business risk, including agentic and autonomous system behavior Accounting Today. It produces control mapping and attestation evidence that looks similar to what a GAO audit would probe. But it is an enterprise risk framework, not a flight-safety one, and it cannot substitute for an airworthiness certification basis that doesn’t yet exist.

Compliance leaders in aerospace should stop treating this as a wait on a rule number. The decision in front of them is whether to start producing ISO 42001 governance records and NIST AI RMF measurement logs now, so that whatever certification basis eventually extends or supplements DO-178C has documentation to certify against instead of a blank page.

---

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 DO-178C’s requirements-based certification model cannot accommodate learned, probabilistic systems and no replacement exists—is technically sound and well-supported, though t
Source & Claim Verification	Qwen · hosted fallback (qwen-2.5-72b-instruct)	cleared. All claims are supported by the provided sources, except for those resting on the unopened ‘Flying Tesla and flying cars’ source.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately identifies the gap in DO-178C for AI certification and correctly contrasts it with ISO 42001, NIST AI RMF, and the EU AI Act without overstating their applicability to airworth
Technical Accuracy	Llama	cleared. The article accurately highlights the limitations of DO-178C in certifying AI components in safety-of-life systems and the need for a new certification basis that incorporates emerging AI assurance fr
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies the core problem of a missing certification basis for AI in aviation and critically assesses existing frameworks, though it could benefit from explicitly stating po
Novelty & Non-Duplication	Grok	held. Restates the repeatedly covered aerospace DO-178C/AI certification gap, FAA-EASA limits, and ‘produce ISO 42001/NIST logs now’ advice, with the ATC pause request as a thin new hook that adds no materi
Validation	DeepSeek	cleared. The briefing establishes a clear and plausible certification gap by contrasting the deterministic nature of DO-178C with the probabilistic nature of AI, which aligns with the rationale for the congres

Sources cited: 13. Validation challenges: 0. Production cost: $0.16 (measured across drafting, revisions and every board review, plus an estimated image). Learn how these briefings are written and verified.