AI-Supported Risk Assessment Without Data Leaving Your Organisation

Risk assessment is a very important step of safe design across sectors, from machinery and robotics to logistics, automotive, and process industries. Yet the complexity of modern systems, combined with increasingly strict standards and regulatory expectations, places high demands on engineering teams. At the same time, organisations are justifiably cautious about using cloud-based AI tools and LLMs (e.g. ChatGPT, Gemini) alike, as safety-relevant hazard data, customer projects, or proprietary machine descriptions must remain confidential.

Our recently published work on a prototype implementation of a local, multilingual AI-based expert system for functional safety demonstrates how risk assessment tasks can be modernised without compromising data protection. Built with a Rasa conversational interface and an expert-curated knowledge base, the system guides users through hazard identification, PLr determination, and risk-reduction strategies in accordance with ISO 12100 and ISO 13849, all running fully on-premises.

As the system is deployed locally, no machine descriptions, photographs, CE files, or customer scenarios leave the organisation. This enables companies in manufacturing, automation, automotive, railway and industrial equipment design to adopt AI-supported workflows while meeting internal confidentiality rules, NDAs, and data-protection requirements.

By combining structured expert knowledge, local NLP models, and transparent reasoning, this approach offers a practical path toward AI-enabled risk assessment that is both standards-aligned and privacy-preserving. It provides a scalable way to support engineers, improve consistency, and reduce assessment time, without exposing sensitive information to external services.

Here is the link to the published work: https://onlinelibrary.wiley.com/doi/10.1111/risa.70151

Learn more about innotec’s risk assessment services here: https://innotecsafety.com/consulting/risk-assessment