In a large-scale IT project in the public sector, we regularly face a challenge that appears simple but is actually complex: comparing a large number of complex PDF documents with HTML pages that have identical content. The documents contain critical technical content that must be checked for exact consistency – a mammoth task that is time-consuming and resource-intensive when done manually. Automating the pre-processing required for this comparison is also extremely challenging. The devil is in the details here, in the form of unstructured layouts, changing formatting, nested tables and multi-column content – and it is also difficult to match the sections of the respective document types that need to be checked.
At this point, we used AI-based methods to support three key tasks: the structural analysis of PDF layouts, the assignment of sections, and intelligent text comparison in the form of automated deviation reports. In doing so, we encountered a key area of tension: how can an LLM be used effectively as a non-deterministic system when, in contrast to typical use cases, maximum sensitivity is required rather than a ‘good average value’ – in other words, every deviation should be detected without false negatives? In this presentation, I will report on the practical challenges of integrating AI into strict QA processes, our approaches to combining classic methods with AI models, and dealing with uncertainty and validation in test-productive use. A field report with lessons learned from practice – between automation, quality and reliability.

Automotive engineering is undergoing a profound transformation. Agentic-AI—autonomous agents capable of reasoning, learning, and acting within controlled boundaries—is emerging as a new actor in the design of safety-critical systems. More than a simple tool, it is becoming a true engineering partner, supporting teams in Functional Safety, Cybersecurity, and Safety Management activities.

This presentation offers a concrete vision of human–AI collaboration, where Agentic-AI helps make engineering processes more coherent, traceable, and adaptive, while meeting the rigorous expectations of ISO 26262, ISO 21434, and ISO PAS 8800. It can accelerate the drafting and traceability of requirements, assist in risk analyses (HARA, TARA), support automated generation of safety arguments (GSN), and facilitate compliance reviews—strengthening process rigor without slowing development.

Far from replacing engineers, Agentic-AI should be viewed as a natural extension of their skills—a new competence that enhances control and efficiency in an increasingly complex technical environment.

To illustrate this potential, a use case will focus on AI-assisted generation and evaluation of safety requirements. Building on INCOSE and ISO 26262-8 criteria, the system detects ambiguities, identifies missing details, and proposes clearer, testable formulations.

This experience shows how Agentic-AI can help build a truly holistic approach to safety, preparing automotive engineering for the era of Software-Defined Vehicles.

This presentation shares Arm’s three-year journey towards stronger efficiency, compliance, and safety culture in automotive — structured around People, Process, and AI/Automation.

• People: Three years ago, when I was appointed Arm’s Global Safety Manager, I asked engineers where to invest first. The clear answer was safety culture. Behind every good DFMEA, FMEDA, every on-time delivery, and every zero-DPPM milestone stand the people. I will share, in an interactive and collaborative format, how we shaped safety culture: from workshops and trainings to safety meters and leadership engagement. What worked, what didn’t, and how to measure safety culture and lead positive change in a diverse global organization.
• Process: With culture established, we could tackle process improvement. Process change takes time — and cannot be dictated top-down. We applied a “crowdfunding” approach to process development, focusing on maintainability, stability, and adaptability in a dynamic organization. Last month, our framework achieved its first UL certification — Arm’s first globally certified and aligned process, covering safety, quality, and security. I will share how we built it, challenges faced, and what comes next. The process integrates ISO 26262, ISO/SAE 21434, ASPICE, and IATF 16949.
• AI and Automation: Only with People and Process in place can we safely integrate AI. AI cannot act as a certified software tool, yet it can still be harnessed effectively. We developed a practical approach using modular, machine-readable templates and an integrated AI reviewer. Our ambitious but realistic target is to reduce manual work by up to 80% — without compromising compliance or safety.

This case study combines practical lessons, interactive elements, and real results to show how culture, process, and technology can together drive measurable impact in automotive quality and safety.

Organizers