Saab's AI Fighter Agents and the Certification Gap Nobody Has Closed

Saab’s AI Fighter Agents and the Certification Gap Nobody Has Closed

Saab has begun flying AI agents in beyond-visual-range trials on the Gripen E, and the company says safety and control were built into the development process from the start. What it has not published is the verification framework, the independent evaluation procedures, or the operational limits governing those agents, leaving open exactly the questions a defense procurement authority or allied regulator would need answered before accepting the aircraft into service, as TheDefenseWatch notes.

This matters because the assurance regime aerospace has relied on for decades, DO-178C, was designed around deterministic software with fixed logic paths that can be traced requirement by requirement to test cases. An AI agent making tactical decisions in a dynamic BVR engagement does not behave that way. Its outputs depend on training data, model weights, and runtime context that shift with each scenario. Saying safety was “integrated from the beginning” is a design philosophy claim, not a certification artifact. Regulators and allied air forces evaluating Gripen E, or any platform fielding similar agents, will need something closer to a documented assurance case: traceable test coverage across the agent’s decision space, defined failure modes, and a basis for determining when the system is operating outside its validated envelope. None of that has been made public, and it is unclear whether it yet exists in a form regulators would accept.

The eVTOL sector offers a preview of how long this kind of gap can persist once novel capability outruns certifiable process. No piloted eVTOL has secured a type certificate from the FAA, EASA, the UK CAA, Brazil’s ANAC, or Japan’s JCAB, despite years of flight testing and repeated program announcements, according to The Aviation Hub. The common thread is not immaturity of the underlying technology. It is that existing certification frameworks were not written for the category now in front of them, whether that category is distributed electric propulsion or adaptive decision-making software, and closing that gap takes longer than any program schedule assumes.

For defense primes and their suppliers, the lesson is that demonstrating an AI agent can fly is a necessary but insufficient signal. The binding constraint will be whether the verification evidence exists in a form an airworthiness authority, or an allied customer’s technical authority, can actually audit. That evidence should be built concurrently with the capability, not retrofitted once trials generate headlines. Programs that wait until an agent is flight-proven to start building the assurance case will find themselves in the same queue as eVTOL developers: technically airborne, administratively grounded.

The organizations positioned to move fastest once a certification pathway for AI-enabled flight systems does emerge will be the ones that treated verification as a parallel workstream from day one, not a deliverable they owe regulators after the fact.

---

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 demonstrating capability without concurrent verification evidence creates a certification bottleneck—is coherent and well-supported by the eVTOL analogy, though the piece would
Source & Claim Verification	In-house seat · hosted fallback (qwen-2.5-72b-instruct)	cleared. The draft accurately cites sources for its claims, but the EASA source is not useful for verification and the eVTOL certification delays are not directly linked to AI certification issues.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately identifies the DO-178C certification gap for AI-driven systems but does not substantively address ISO 42001, EU AI Act, NIST AI RMF, or other sector-specific frameworks (HIPAA,
Technical Accuracy	Llama	cleared. The article accurately identifies the certification challenges associated with AI agents in aerospace, highlighting the limitations of DO-178C and the need for a documented assurance case.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and critiques the lack of public counterarguments or detailed verification frameworks from Saab, highlighting a critical gap in certification readiness rather than
Novelty & Non-Duplication	Grok	held. Restates the Nov 2–3 aerospace/DO-178C assurance-gap thesis (and eVTOL parallel) with Saab Gripen trials as a fresh hook but no new certification fact, framework, or decision for the reader.
Validation	DeepSeek	cleared. The central claim is plausible and not refutable, but it leans on an absence-of-evidence assertion that Saab has not published any verification framework, which the supplied sources do not independent

Sources cited: 8. Validation challenges: 0. Learn how these briefings are written and verified.