Foreword

Eventhough this blog is talking about drones, we believe it is very relevant to mention that EASA is also working on trustworthy AI for all aviation sectors. Soon this regulation will also impact unmanned aviation, especially when it comes to more autonomous drones and UAS traffic management systems.

Overview

Artificial Intelligence (AI) is transforming aviation, promising safer skies and smarter operations. But trustworthiness is non-negotiable. The EU AI Act (Regulation (EU) 2024/1689) sets strict requirements for high-risk AI systems, and EASA is leading the charge with NPA 2025-07, introducing the DS.AI framework for AI trustworthiness in aviation. This blog explores:

• ▸
  Why AI trustworthiness matters.
• ▸
  How the EU AI Act shapes aviation.
• ▸
  EASA’s vision and leadership through Rulemaking Task RMT.0742.
• ▸
  Detailed insights into DS.AI articles: Scope (DS.AI.010), Compliance (DS.AI.040), Risk Assessment (DS.AI.130), and Human-Centered Design (DS.AI.170).
• ▸
  Practical examples of AI in cockpits, maintenance, and air traffic control.
• ▸
  Future-proofing for generative AI and reinforcement learning.
• ▸
  How stakeholders can shape the future during the public consultation open until February 2026.

Glossary of Acronyms

• ▸
  AI – Artificial Intelligence: Technology that infers from inputs to produce outputs like predictions or decisions.
• ▸
  EASA – European Union Aviation Safety Agency: The EU body responsible for aviation safety.
• ▸
  NPA – Notice of Proposed Amendment: A formal document proposing regulatory changes for public consultation.
• ▸
  DS.AI – Detailed Specifications for AI Trustworthiness: Technical requirements introduced by EASA.
• ▸
  AMC – Acceptable Means of Compliance: Non-binding guidance to meet regulatory requirements.
• ▸
  GM – Guidance Material: Additional explanatory content supporting compliance.
• ▸
  CRT – Comment-Response Tool: Platform for submitting feedback on NPAs.
• ▸
  AL – Assurance Level: A measure of confidence in AI system safety.
• ▸
  OD / ODD – Operational Domain / Operational Design Domain: Conditions under which an AI system is intended to operate.
• ▸
  EU AI Act – Regulation (EU) 2024/1689: The European Union’s legal framework for AI.

Introduction: Why Trustworthy AI Matters

Imagine a cockpit where AI predicts turbulence before the pilot sees it, or an air traffic control center where AI optimizes flight paths in real time to reduce delays and emissions. These scenarios are no longer science fiction—they're on the horizon. But in aviation, innovation cannot come at the expense of safety. Trustworthiness is the bedrock of every system that interacts with human lives.

Quick Insight:

AI systems are probabilistic, meaning they learn from data and make predictions rather than following fixed rules. This flexibility is powerful—but it introduces uncertainty. In aviation, uncertainty must be managed with precision.

The Regulatory Imperative: EU AI Act and Aviation

EU AI Act – Chapter III, Section 2

The EU AI Act (Regulation (EU) 2024/1689) is the world’s first comprehensive legal framework for AI. Chapter III, Section 2 sets mandatory requirements for high-risk AI systems, including:

• ▸
  Article 8: Compliance with requirements.
• ▸
  Article 9: Risk management system.
• ▸
  Article 10: Data and data governance.
• ▸
  Article 11: Technical documentation.
• ▸
  Article 12: Record-keeping.
• ▸
  Article 13: Transparency and provision of information.
• ▸
  Article 14: Human oversight.
• ▸
  Article 15: Accuracy, robustness, and cybersecurity.

Did You Know?

High-risk AI systems include those used in aviation, healthcare, and critical infrastructure. These systems must meet strict safety and transparency standards before deployment.

Article 108: Aviation-Specific Obligations

Article 108 amends the Basic Regulation (EU) 2018/1139, integrating AI requirements into aviation law. This means:

• ▸
  AI systems in aviation must meet trustworthiness standards.
• ▸
  Certification and approval frameworks must adapt to AI’s unique characteristics.
• ▸
  Human oversight remains essential—AI cannot replace accountability.

EASA’s Vision and Leadership

EASA’s Rulemaking Task RMT.0742 operationalizes the AI Act for aviation. Its objectives:

• ▸
  Ensure AI trustworthiness for safe use in aviation.
• ▸
  Enable AI deployment in domains identified in Article 108.
• ▸
  Prepare for future AI technologies, including generative AI and reinforcement learning.

Quick Insight:

EASA's roadmap includes two major steps:

2025: NPA 2025-07 introduces a generic AI trustworthiness framework.

2026: Second NPA adapts this framework to domain-specific regulations (airworthiness, operations, ATM/ANS).

Deep Dive into DS.AI Framework

DS.AI.010 – Scope and Exclusions

DS.AI applies to AI-based systems classified as:

• ▸
  Level 1: Human augmentation or assistance.
• ▸
  Level 2: Human-AI cooperation or collaboration.

Excluded:

• ▸
  Systems with catastrophic failure potential.
• ▸
  Online learning systems.
• ▸
  Hybrid AI with high safety impact.
• ▸
  AI verification tools without independent human checks.

Example:
An AI tool that assists pilots with weather predictions (Level 1) falls under DS.AI.010. A fully autonomous flight control system with catastrophic failure potential does not—it’s excluded for now.

DS.AI.040 – Compliance Process

Organizations must:

Define processes to meet DS.AI requirements.
Document technical compliance, including risk assessments and human oversight measures.

Quick Insight:

This aligns with Article 11 (Technical Documentation) and Article 12 (Record-Keeping) of the AI Act.

Example:
An aircraft manufacturer integrating AI for predictive maintenance must maintain detailed documentation of algorithms, data sources, and validation results.

DS.AI.130 – Risk Assessment

Risk assessment is the cornerstone of trustworthiness:

• ▸
  Hazard Classification: H1 (fatalities) to H5 (no impact).
• ▸
  Likelihood: Frequent to extremely improbable.
• ▸
  Assurance Levels (AL): AL2–AL6 based on risk severity.
• ▸
  Continuous monitoring and mitigation.

Did You Know?

Hazard likelihood is expressed per operational hour. For example, "extremely improbable" means less than 1 in a billion hours.

Example:
An AI system for automated taxiing might be classified as H3 (significant performance degradation risk) and require AL3 assurance, including rigorous testing and fallback procedures.

DS.AI.170 – Human-Centered Design

AI must:

• ▸
  Provide operational explainability (Article 13).
• ▸
  Support shared human-AI situational awareness.
• ▸
  Enable dynamic task allocation and error management.

Quick Insight:

Explainability means the AI system can tell the operator why it made a recommendation—crucial for trust and accountability.

Example:
In air traffic control, an AI assistant suggesting conflict resolutions must explain its reasoning clearly and allow controllers to override decisions instantly.

Human-AI Teaming in Action

EASA introduces AI levels:

• ▸
  Level 1A: Information augmentation (e.g., AI-enhanced weather radar).
• ▸
  Level 1B: Decision-making support (e.g., route optimization suggestions).
• ▸
  Level 2A: Directed automatic decisions (e.g., autopilot adjustments under supervision).
• ▸
  Level 2B: Supervised collaboration (e.g., AI managing multiple ATC tasks with human oversight).

Did You Know?

Human oversight is non-negotiable. AI is a partner, not a replacement.

Future-Proofing Aviation AI

Today’s DS.AI focuses on supervised and unsupervised learning. Tomorrow’s challenges include:

• ▸
  Reinforcement learning for adaptive flight control.
• ▸
  Hybrid AI combining symbolic reasoning and machine learning.
• ▸
  Generative AI for predictive modeling and simulation.

EASA anticipates these trends and aligns with EUROCAE WG-114 / SAE G-34 standards for learning assurance.

Benefits and Challenges

Benefits

• ▸
  Harmonized framework across aviation domains.
• ▸
  Safer integration of AI into critical operations.
• ▸
  Flexibility for innovation.

Challenges

• ▸
  Initial compliance costs.
• ▸
  Technical immaturity of assurance methods.
• ▸
  Balancing safety with rapid technological evolution.

Call to Action: Public Consultation

EASA invites stakeholders to comment on NPA 2025-07 via the Comment-Response Tool (CRT) by 10 February 2026. Your feedback will shape the next phase of AI regulation in aviation.

Closing Vision: A Safer, Smarter Aviation Ecosystem

EASA is not just following the AI trend—it’s leading the way. By embedding trustworthiness into the DNA of aviation AI, EASA ensures that innovation never compromises safety. Together, we can build an ecosystem where AI empowers humans, enhances safety, and earns public trust.

Key Takeaways

• ▸
  AI in aviation is here to stay—but trustworthiness is essential for safety and public confidence.
• ▸
  The EU AI Act (Regulation (EU) 2024/1689) sets strict requirements for high-risk AI systems, including risk management, transparency, and human oversight.
• ▸
  EASA’s NPA 2025-07 introduces the DS.AI framework, a flexible and robust foundation for AI trustworthiness in aviation.
• ▸
  DS.AI highlights:
• ▹
  DS.AI.010: Defines scope and exclusions for AI systems.
• ▹
  DS.AI.040: Establishes compliance processes and documentation requirements.
• ▹
  DS.AI.130: Introduces structured risk assessment and assurance levels.
• ▹
  DS.AI.170: Emphasizes human-centered design and explainability.
• ▸
  Human-AI teaming is the future—AI will assist, not replace, human decision-making.
• ▸
  EASA is future-proofing regulations for reinforcement learning, hybrid AI, and generative AI.
• ▸
  Public consultation is open until February 2026—your feedback will shape the next decade of aviation innovation.

References

• ▸
  EU AI Act (Regulation (EU) 2024/1689): Articles 8–15, 108.
• ▸
  EASA NPA 2025-07 (A & B): DS.AI.010, DS.AI.040, DS.AI.130, DS.AI.170.
• ▸
  Terms of Reference RMT.0742.