EASA Artificial Intelligence Days 2025: Highlights of Day 2

Day 2 of the EASA Artificial Intelligence Days 2025 focused on use cases, innovation, and research, featuring presentations from industry and research organizations, as well as panel discussions on human factors and the qualification of AI-based operational tools.

If you have missed it, in the EASA Artificial Intelligence Days 2025: Highlights of Day 1 article, you will find out the key highlights of Day 1

Let’s dive in! 🤿

Day 2 emphasized practical applications of AI through Innovation Partnership Contracts (IPCs), research projects, and panel discussions.

Key themes included the critical role of human factors, the cautious but forward-looking use of Generative AI, and the need for proportional qualification of AI-based tools.

1. IPC – BEACON Project (Boeing - EASA)

• Launched in 2023, this IPC explored ways to certify a Level 2 machine learning application for transport aircraft (CS25/FAR25).

• Focus areas included Level 2A auto taxi applications, Level 3 capabilities, and the Safe Runway system for detecting intruders before landing.

• A live demonstration of both systems took place at a Boeing facility, marking a major milestone.

• Key discussions stressed tailoring objectives to each system, rethinking the definition of “end user,” and clarifying AI classification impacts.

3. IPC - Generative AI for Technical Operations (Lufthansa Group - EASA)

• Focused on using AI to enhance maintenance troubleshooting, safety, and efficiency.

• Key approach: a RAG architecture to ground AI answers in validated company data, supported by a “verification agent.”

• Classified as Level 1B (human-assisted, human responsible).

• Challenges include defining ODDs, operational explainability, and regulatory guidance for Gen AI.

4. Panel Discussion: AI-Based Operational Tools and Qualification (Airbus Protect; SixFoisSept; Swiss International Airlines; EASA)

• Addressed AI tools across production, design, maintenance, ATM, and authority oversight.

• EASA outlined a risk-based assessment with three gateways:

  • impact on approved activities

  • independent verification,

  • and contribution to safety risk.

• It was highlighted the challenges with incomplete requirements, blackbox behaviors, and the need to qualify verification tools.

• It was showed AI’s potential for oversight by prioritizing safety-critical inspections.

• Other discussions emphasized user involvement, staff training, and managing fast-changing LLM lifecycles.

5. SESAR Projects

• Jarvis Project (Collins): AI-based digital assistants for ATC and airports, with deep learning for communication and ML for runway predictions. Challenges included accuracy, explainability, and interface design.

• Darwin Project (Honeywell): Explored human-AI teaming and pilot state monitoring. Addressed ATC communication errors and HMI complexity, stressing representative environment testing.

• SynthAIR Project (SINTEF): Developed generative models for synthetic ATM data, enabling safer experimentation, protecting privacy, and reducing bias. Applications included delay prediction, diversion analysis, and filling data gaps.

6.Closing Reflections

Day 2 of the EASA AI Days 2025 brought the conversation closer to practice, demonstrating how research and innovation can evolve into real-world applications.

It was clear that progress in AI is happening, but not without its challenges.

The recurring thread across sessions was balance: between ambition and caution, innovation and oversight, speed and safety.

EASA’s reaffirmed a pragmatic, risk-based, and ethical approach.

From the Center…

Looking beyond aviation, I’m reminded that innovation always tests the boundaries of what we know and trust.

The real challenge is not just developing powerful tools, but shaping the conditions for their safe and meaningful use.

See you next week 👋

Disclaimer: The information provided in this newsletter and related resources is intended for informational and educational purposes only. It reflects both researched facts and my personal views. It does not constitute professional advice. Any actions taken based on the content of this newsletter are at the reader's discretion.