"AI makes things up. I won't trust it with airworthiness."
I hear some version of this from engineers often, and they're right to say it. AI models do hallucinate. They can produce a fluent, confident answer that isn't supported by the source in front of them. In most industries that's an embarrassment. In ours, a confident but wrong "not applicable" on an Airworthiness Directive is a finding waiting to happen.
So let me start by agreeing with the sceptics. Then I'll explain why I still think AI belongs in continuing airworthiness, and the conditions under which it does.
Why the scepticism is right
An Airworthiness Directive is a legal instrument. Whether it applies to an aircraft depends on details that are easy to get wrong: the type and model, the serial number range, the engine fitted, the modifications and service bulletins embodied, the directive it supersedes, and its counterpart from the other authority.
Applicability is a chain of assertions, and every link has to hold: registration, serial number, type certificate, model and variant, serial effectivity, engine or propeller, and modification status. Every verdict, whether applicable, not applicable or cannot determine, needs evidence. We've set out the full chain in the Aviagent handbook: How applicability is actually determined.
A hallucination in this context doesn't look like nonsense. It looks like a plausible answer:
- an aircraft marked "not applicable" because a serial-number condition was misread
- a compliance time that doesn't match the directive's actual wording
- a superseded directive treated as current, or the reverse
Each of these reads perfectly well. That's what makes it dangerous. An engineer who distrusts a tool that can be fluently wrong is doing their job.
Why keeping AI out isn't the answer
I don't think the answer is to keep AI out of continuing airworthiness. I think it's as inevitable here as electricity and the internet were. Both were risky when they arrived, and aviation didn't refuse them. It put rules around them.
We've done this before, inside our own industry. The autopilot didn't remove the pilot. Fly-by-wire didn't remove the engineer. Each technology arrived with defined limits, independent verification, and a named human accountable for the outcome. That's why they earned trust. Nobody trusted them because they were clever.
AI should come into CAMOs the same way: under engineering control, not around it. The question isn't whether AI arrives in continuing airworthiness. It's whether it arrives on terms that engineers and Accountable Managers can defend to an auditor.
What “under engineering control” means in practice
That's the principle Aviagent is built on, and it comes down to three rules.
AI does the reading, not the deciding. It turns directive text, cross-references and compliance times into a structured draft that answers the questions an engineer asks anyway. Does it apply to my fleet? When must I comply? What do I do? How often? What happens if I don't?
Every answer stays a draft until a qualified engineer approves it. The engineer reviews it, corrects it where it's wrong, and approves it. Until then it's labelled as a draft. Under EASA Part-CAMO, airworthiness decisions rest with the organisation's qualified personnel. That's the requirement, and it's the right one.
Every draft points back to the directive it came from. The reviewer checks the draft against the source text rather than taking it on trust. A draft you can't check is worth nothing, however good it sounds.
For the Accountable Manager, the result is a determination that is documented, traceable and ready to defend, made by their own engineers.
Where the hours actually go
If AI only does the reading, why use it at all?
Because the reading is where the hours go. Finding an AD takes minutes. Deciding whether it applies to a specific aircraft, and recording why, can take days. EASA and the FAA publish new directives every week, and each one has to be assessed against every aircraft it might affect.
What changes is where the engineer's time goes: less time assembling the evidence, more time judging it. AI shortens the reading. It doesn't sign anything.
Start with something you can check
You can start without any AI at all. AD Lookup is our free reference tool: type a registration and see the directives that apply to it.
- 46,000+ EASA and FAA directives, synced daily
- 100+ national registers plus the US registry
- 17,000+ EASA–FAA cross-references, so a directive met under one authority isn't counted twice
Five free lookups, no signup. Check the results against your own records. That's the point.
Scepticism is a safety skill
Twenty-five years in continuing airworthiness at SAS, GE Aviation, Pratt & Whitney and HAL taught me that scepticism is a safety skill, not an obstacle to progress. I don't want engineers to trust AI. I want them to be able to check it.
If you're an engineer: what would an AI tool have to show you before you'd let it draft an applicability assessment? I'd genuinely like to hear it. Get in touch.
AD Lookup is reference data, information only, and not a substitute for your CAMO's records. Applicability determinations and sign-offs remain with qualified personnel.