EASA's AI Olympics Closed Early — and That's Actually the Point
The entry window for EASA’s inaugural AI Olympics closed on July 7 — two days ahead of the official deadline — because the agency hit its self-imposed cap of 50 submissions. EASA announced it was no longer accepting entries as the maximum permissible number of 50 had been received, and thanked all entrants while proceeding with evaluation. The top ten will earn a stage at the 4th edition of EASA AI Days, hosted at EASA’s premises in Cologne on September 9–10, 2026.
That’s a small logistical detail that carries a larger signal: the industry has more aviation AI to show than EASA has bandwidth to review.
I think that gap — between the volume of AI being built for aviation and the institutional capacity to evaluate, validate, and eventually certify it — is the most important tension in flight operations technology right now. Not the algorithms. Not the data. The evaluation infrastructure.
The Showcase Problem
The AI Olympics format is genuinely interesting as a regulatory instrument. The conference will focus on latest developments in AI for aviation, including updates on EASA’s concept paper on AI applications and ongoing rulemaking activities, with discussions on AI assurance, human factors, and ethics-based assessment, alongside presentations from industry on reinforcement learning, large language models, and knowledge-based AI technologies. EASA is opening the conference to the best aviation industry AI cases — a novel opportunity for selected entrants to present their practices to all participants.
What EASA is really doing here is creating a structured channel for the regulator to see what the market has actually built. That’s smarter than it might look from the outside. The aviation AI vendor landscape has fragmented quickly, and most of what’s being developed sits somewhere between proof-of-concept and limited airline deployment — useful enough to sell, not yet well-understood enough to certify. A showcase forces vendors to articulate their systems in terms a regulator can evaluate. That friction is productive.
But here’s what I keep coming back to: fifty entries, ten slots, one two-day conference. The math on this suggests a lot of interesting aviation AI is going to remain invisible to EASA’s rulemaking process for another year at least. And EASA’s rulemaking process matters enormously for anyone building AI that touches flight operations — because certification pathways don’t exist yet for most of what the industry is trying to deploy.
The Adoption Problem Is Upstream of the Tech
I recently polled industry professionals on the question of where AI in the cockpit should go first. The results split almost cleanly in three: 40% said maintenance and documents, 20% said pilot decision support, and 40% said AI should be kept out of the cockpit entirely. That last number is striking. Four out of ten respondents — people who work in aviation technology — aren’t ready to put AI anywhere near flight operations in an active role.
That doesn’t surprise me, and I don’t think it reflects technophobia so much as a rational response to where aviation has always been culturally. The industry is heavily dependent on legacy systems and tried-and-true methods, and breaking free of those — even before AI entered the conversation — was never easy. The path forward with AI will run through reliability and trust: operators and regulators both need to feel a demonstrable net advantage before adoption makes sense to them, and right now the certification framework to support that confidence doesn’t fully exist yet.
Hitting 50 entries before the deadline closed is, I think, genuinely both a sign of ecosystem maturity and a sign that vendors are hungry for any regulatory visibility they can get. I’d expect the entrant list to reflect that mix — large legacy vendors who have been working toward this kind of engagement for years, alongside startups whose roadmaps depend on getting in front of regulators early. Those aren’t mutually exclusive motivations.
From my own years working on bid processes and product delivery in this space, I’ve watched vendor roadmap decisions get driven not by what’s technically possible but by what a regulator is willing to engage with. Vendors don’t invest heavily in capabilities they can’t eventually get approved. But a showcase like this does more than validate what vendors were already building — it helps shape the regulatory framework itself, which in turn helps loosen the timeline constraints that vendors are planning around. That’s a meaningful feedback loop. In my experience, vendors are genuinely eager to work with regulators on initiatives like this, not just present to them.
What September Will Tell Us
The ten finalists who present in Cologne in September will effectively become a proxy for where the European aviation AI industry thinks it is. That’s worth watching closely, because the categories of AI that make it to the stage will reveal something about what vendors believe EASA is actually ready to hear about.
My working assumption is that the finalists will skew toward lower-stakes applications — flight data analytics, document processing, weather interpretation — rather than anything that sits directly in the decision loop on an active flight. Realistically, those simpler, lower-risk tasks are where the first steps toward AI in aviation should land, and most vendors building for certification know that. But I’d also expect a few bolder entries in the mix, and I think those are actually valuable in a different way: they help sketch out what the regulatory framework will eventually need to account for, and they give EASA a concrete look at what lies ahead rather than just what’s ready to ship today.
The harder version of the question, which September probably won’t answer, is how quickly EASA’s ongoing rulemaking work under RMT.0742 translates into real guidance that vendors and airlines can use to design certifiable systems. That rulemaking task is progressing in stages, with a second NPA expected in 2026 to propagate the AI trustworthiness framework into domain-specific regulations. A showcase like this is genuinely useful — it gives regulators a clearer picture of what’s happening in the industry and how far along vendors actually are, which should make the conversations between both sides faster and more grounded. But the showcase is a pressure valve. The rulemaking is the actual work, and no amount of stage time in Cologne substitutes for the second NPA arriving with something vendors can build to.
I’m genuinely curious which categories of flight-ops AI dominate the finalist list — and whether any of the ten systems presented in Cologne are from startups rather than established vendors. That ratio would tell us a lot about who EASA’s evaluation criteria are actually designed for.