IN Brief:
- CIBSE published its AI position statement on 9 November 2026 following work by its AI Working Group.
- The institution says engineers remain accountable for outputs, which must be critically assessed and verified before use.
- The 42-page statement addresses professional governance, legal exposure, confidentiality, skills, and future review of guidance.
The Chartered Institution of Building Services Engineers (CIBSE) has published a position statement on artificial intelligence in building engineering, requiring continued professional judgement and critical verification of AI-generated technical work. Issued on 9 November 2026, the guidance was developed by its AI Working Group and supports responsible adoption of tools used in design and operation. Engineers remain accountable for their outputs even when software assists with calculations, documentation, or decisions affecting building performance and safety.
The 42-page CIBSE AI Position Statement examines the use of AI within engineering practice, including professional responsibility, competence, governance, and safeguards for technical work. Its publication does not create a new Act of Parliament or alter existing statutory duties. Rather, it describes the institution’s position on how AI tools should be used alongside the obligations already attached to professional engineering decisions. The institution intends to review its approach as technology, regulation, and professional practice develop.
AI tools can assist with information retrieval, documentation, data handling, and the analysis of complex design options, but those applications require different levels of scrutiny. An automatically drafted meeting note does not carry the same risk as a ventilation or overheating calculation used to support a design decision. The reliability of a result depends on the completeness of its inputs, the suitability of the method, and whether assumptions are visible to an engineer who can test them. CIBSE’s distinction between assistance and professional judgement follows from those differing consequences.
Building services design involves linked calculations of heating and cooling loads, ventilation rates, electrical demand, equipment characteristics, and control conditions. An AI-generated answer can appear credible while relying on a wrong input unit, an unsuitable operating condition, or an unsupported assumption. Checking a result therefore involves tracing its inputs, underlying calculations, and references against the project’s design criteria, not simply editing fluent prose. CIBSE’s requirement for qualified human verification is particularly relevant where a mistake could affect occupancy conditions, energy use, or safety.
Confidentiality can constrain the use of externally hosted tools. Engineering practices hold client specifications, design models, operational data, and commercially sensitive documents, some of which cannot be placed into a third-party service without checking contracts and information-governance policies. The CIBSE position considers the responsibilities accompanying AI use but does not prescribe one software product or a universal arrangement for data retention. Organisations must determine appropriate controls for the information and systems they use, including what may be uploaded and who may access outputs.
The position also addresses professional competence as automation becomes more capable. Engineers who delegate routine analysis to software still need the underlying understanding to identify physically impossible results and ask for better evidence. That requirement becomes acute when designers are presented with AI-generated calculations, specifications, or compliance narratives without a traceable derivation. Training and reskilling therefore concern more than tool familiarity: the ability to interrogate calculations, evaluate uncertainty, and recognise scope limits remains central to accountable design.
Agentic AI, which can undertake sequences of tasks rather than respond to a single prompt, expands the possible effect of an error. A system that extracts information, proposes options, and updates documents may propagate an early incorrect assumption into several downstream outputs. Appropriate human checkpoints and records of what changed are consequently important considerations. The CIBSE document examines agentic systems and the boundary of engineering judgement, but it should not be presented as certifying that any particular autonomous workflow is suitable for safety-critical use.
The engineering documents used during construction connect design decisions to procurement, installation, commissioning, and eventual operation. If AI introduces an error into a plant schedule or a technical specification, the mistake can move through those stages before it is found. Document revision controls, design checking, and clear approval responsibilities remain necessary regardless of how a document was produced. CIBSE’s position retains those established professional obligations rather than presenting AI as a separate approval route.
CIBSE cites its Code of Conduct and engineering principles as part of the framework for responsible use. The institution also plans to review its position as the capability of AI systems, professional applications, and regulation evolve, rather than treating the 2026 document as a fixed final standard. Its knowledge portal identifies the publication as active and freely available. The existence of other industry guidance, including for surveyors and architects, does not make those separate documents interchangeable with engineering duties under CIBSE’s professional framework.
CIBSE intends its statement as the beginning of an evolving professional position and will review it as AI applications and regulation change. Its immediate practical requirement is critical assessment by suitably qualified engineers before AI-generated content informs technical work. Professional responsibility remains with the engineers and organisations using the systems, including where an automated output contributes to a building design or operational decision.
