When the Algorithm Reads Regulations
How the Bern Building Insurance Company Uses AI in Fire Protection—and What That Means for Planners, Authorities, and the General Public
Fire safety regulations are complex. Often incomprehensible to laypeople and time-consuming for professionals—and with the comprehensive revision of the fire safety regulations in 2026, which is expected to take effect in 2027, things aren’t getting any easier. David Sauser, Head of Prevention and Intervention at Gebäudeversicherung Bern (GVB), therefore began wondering early on whether artificial intelligence could help clear a path through the regulatory jungle. The answer is AI4Fire: a web-based tool that assists architects and building designers in preparing fire safety documentation.
From Forum to AI Platform
The journey began with a recurring problem: «We received many inquiries from citizens who weren’t quite sure how to deal with fire safety regulations,» Sauser explains. In 2015, the GVB therefore launched a fire safety platform—initially with a forum designed to provide standard answers to the most common questions. «The basic idea was: We need something that provides answers so we don’t have to answer the same question seven times.»
This led to the creation of Heureka Plus—a platform that automatically maps prescriptive codes. «We wanted to replace the traditional »if-then’ queries with decision trees: The user enters information about their building and is told which codes apply to it.”
During user testing, it quickly became clear: the devil is in the details. «Visuals beat text,» says Sauser, summarizing one of the key findings. «We had to keep it brief, use clear language, and explain technical terms in a way that laypeople could understand. And: Users need to experience their first sense of accomplishment after no more than twenty questions—otherwise, they’ll give up.»
The Great Disillusionment
When ChatGPT became available to the general public in November 2022, there was a great deal of excitement. GVB invested in the technology, tested it, and adapted the model to its own needs. In the process, they encountered a fundamental problem: the AI’s hallucinations. «Sometimes the result was incoherent.»
Specifically: A requirement such as «EI30» may be mandated for various reasons—for example, to limit the size of a fire compartment or in connection with hazardous materials. «That can confuse the AI and lead to nonsensical results,» says Sauser. For a tool intended to serve as the basis for a building permit, that’s not an option.
As a result, the GVB continued to rely on the tried-and-true target trees for the actual fire safety verification for the time being, while continuing to refine its approach in parallel.
AI4Fire – The Escape Plan from the Maze of Regulations
AI4Fire emerged from this process. The tool verifies compliance with fire safety regulations in real time, and an integrated GPT model answers questions clearly and comprehensibly. «Architects can clear up potential obstacles early on. The result is a pre-reviewed certification that streamlines the building permit process and provides planning certainty,» explains Sauser.
At the same time, the technology has advanced rapidly. «The growth has been exponential. The tools have improved dramatically in a very short time—we’ve definitely noticed that.» The problem of hallucinations, which plagued the early attempts, is much less of an issue today.
BSV 2026 and the Next Step
With the complete revision of the Fire Safety Regulations 2026, the industry faces a new challenge. The new Fire Safety Regulations are deliberately written like a legal text—with cross-references, a consistent structure, and no repetitions. «We feel that this will be difficult for many non-experts,» says Sauser.
Other building insurance companies took notice: GVZ reached out, and the Aargau and Vaud building insurance companies followed suit. Together with the VKF, the four cantonal building insurance companies are now using the POC to develop a market-ready product that will be available when the BSV launches in 2026.
Responsibility remains with people
Despite all the enthusiasm, Sauser sets clear boundaries. He says AI is always just a snapshot in time. The technology will take over repetitive tasks and the drafting of major concepts—«but the responsibility always lies with humans.»
«It takes the brainpower of someone who doesn’t just come up with good concepts in theory, but ones that can actually be put into practice in everyday life.» AI makes work easier, but it doesn’t replace good judgment. In the end, buildings where people live always need someone to take responsibility.
