AISS

The ESReDA Project Group on Artificial Intelligence in Safety and Security (AISS) focuses on the growing intersection between artificial intelligence (AI) and safety and security, addressing both protection against malicious acts (security) and industrial and occupational safety (safety). The group adopts a unified perspective, recognising that safety and security can no longer be treated as separate domains when analysing the risks associated with advanced digital and AI-based technologies.

This Project Group explores how AI and other emerging technologies can strengthen safety and security systems, while also examining how these technologies impact risk prevention, safety assurance, and protection frameworks in modern societies. The focus is explicitly technical and industrial, rather than addressing speculative or existential risks of AI.

The AISS Project Group does not aim to develop new AI tools or algorithms. Instead, its primary goal is to create a structured “toolbox” of principles, guidelines, and best practices to support authorities, regulators, and organisations in the responsible and informed use of AI in safety- and security-critical contexts.

The PG adopts a pragmatic and application-oriented approach, concentrating on:

• Understanding the added value of AI in safety and security applications
• Identifying and characterising the new risks and vulnerabilities introduced by
  AI-based systems
• Supporting decision-makers in balancing innovation with robustness,
  transparency, and trust

The work of the AISS Project Group is structured around five critical thematic areas:

1. Integrity in System Design

   Analysis of how AI-based tools may encourage engineers to move prematurely towards detailed models (“design paradox”), potentially limiting error detection, reducing transparency, and constraining the exploration of alternative design
   options.

2. Accident and Incident Investigation
   Assessment of the potential of AI for analysing Big Data, supporting advanced simulations and 3D reconstructions, while addressing risks related to model hallucinations, biased datasets, and misleading conclusions in accident investigations.
3. Human Competencies and “Informacy”
   Examination of the skills required for professionals to critically interpret AI outputs, moving beyond digital literacy towards a deeper competence that preserves human judgement, intuition, and responsibility in safety-related decision-making.
4. AI as a “Whistleblower”
   Exploration of AI systems as impartial and anonymous mechanisms for detecting non-compliance with procedures or unsafe practices, including the ethical, organisational, and privacy-related implications of such applications.
5. Regulation, Certification, and Assurance
   Analysis of how regulatory frameworks must evolve towards continuous certification and adaptive auditing, recognising that AI systems learn, evolve, and change their behaviour over time.