EASA AI Concept Paper (Issue 02) and learning assurance

Why this page is here

The request for this directory named the FAA, SAE G-34 and DO-178C. EASA’s concept paper is the most detailed public aviation-authority guidance on machine learning, and both the FAA roadmap (which says EASA’s work helped inform the FAA roadmap) and SAE G-34 work alongside it.

What it is

EASA published Issue 2 of its Concept Paper on AI and ML on 6 March 2024. It refines guidance for Level 1 AI applications, those enhancing human capabilities, and provides guidance for Level 2 systems built on human-AI teaming.

The paper says it presents a first set of objectives to anticipate future EASA guidance and requirements, and that it does not constitute final detailed guidance. It covers Level 1 and Level 2 AI applications but not yet Level 3.

Who it applies to

Issue 02 is guidance for Level 1 & 2 machine-learning applications in aviation. It is not a US FAA rule. The FAA has said EASA’s work helped inform the FAA roadmap; that is influence, not adoption.

Proposed Issue 03 (consultation, not published guidance)

On 3 June 2026 EASA released Proposed Issue 03 for a 10-week consultation, with comments due no later than August 12, 2026. EASA calls this the final Concept Paper deliverable foreseen under the EASA Artificial Intelligence Roadmap 2.0. The news item says proposed Issue 03 explores Level 3 AI applications, corresponding to ‘advanced automation’ and addresses additional AI techniques, including reinforcement learning and symbolic AI. A “final Concept Paper deliverable” on that news item is not an adopted AMC. The consultation PDF is proposed Issue 03. Level wording from the draft is on EASA AI Concept Paper levels.

NPA 2025-07 (rulemaking, separate from the concept paper)

NPA 2025-07 is dated 10 Nov 2025, with comments through 10/02/2026. It proposes a new set of detailed specifications on artificial intelligence (AI) trustworthiness in response to the Artificial Intelligence Act Chapter III, Section 2. It is a proposal, not an adopted detailed specification. Dates next to the FAA roadmap sit on FAA and EASA AI certification timelines.

Learning assurance

The paper proposes a new concept of learning assurance to extend the traditional development assurance, with the goal of confidence that trained models can generalize with adequate performance on unseen operational data and are robust in foreseeable conditions. It illustrates the process as a W-shaped cycle that adapts the typical development assurance V-cycle to ML concepts, running concurrently with the traditional V-cycle for non-ML constituents.

The paper places an emphasis on data management aspects and learning processes and bases learning assurance on three separate and independent data sets for training, validation and testing. Its guidelines cover supervised and unsupervised learning, but not other types such as reinforcement learning. It ties data quality to the operational design domain: a correct and complete definition of the ODD is a prerequisite to an adequate level of quality of the data sets.

Criticality limits

With supervised learning, EASA says experience is limited and some anticipated means of compliance for the highest criticality are not yet available, so it will initially accept only applications where AI/ML constituents do not include IDAL A or B / SWAL 1 or 2 / AL 1, 2 or 3 items.

Research behind it

EASA’s MLEAP research project aims to identify concrete means of compliance with the learning assurance objectives, covering generalisation guarantees, data completeness and representativeness, and algorithm and model robustness.

Relationship to the EU AI Act

The EU AI Act amends the EASA Basic Regulation: Article 108 requires that, when adopting implementing and delegated acts on AI systems that are safety components, the requirements in Chapter III, Section 2 of the AI Act shall be taken into account. See the EU AI Act page. The supplier brief is EASA Issue 2 explained for suppliers.