Aipedia: Building services engineers
MEP designers should never use AI-written calculations to decide whether an existing roof can carry new plant; that is a structural question for a competent structural engineer. CROSS-UK Report 1554 shows how convincing but wrong AI output reached a real project.
Part of: AI in building services engineering
In short
CROSS-UK Report 1554 (16 July 2026) describes AI-generated calculations used to justify heavy plant on the roof of an existing concrete-framed building. The report looked detailed to a non-engineer but contained errors, with new ones appearing in revisions. The reporter found the roof could carry only light plant. The case involved an architect, but plant weights, locations and supports usually come from the building services designer. CIBSE Code clause 4.2 and Part 2A regulation 11F require work within competence. PI usually responds to negligent design claims, but work outside your professional discipline may raise cover questions.
In CROSS-UK Report 1554, published on 16 July 2026, a new retail tenant in a building of under seven storeys, with a reinforced concrete frame and flat slabs, wanted to install heavy plant on the roof. An architect relied heavily on AI-generated calculations to justify the loading without verification by a competent structural engineer. The reporter, a structural engineer appointed by the landlord, found the roof could support only light plant. The calculation report appeared "robust, detailed, and well written" to a non-engineer, but contained factual errors, and new errors appeared each time it was rewritten after comments.
The CROSS expert panel noted that AI outputs can seem "highly plausible and authoritative" while being wrong, that large language models are not inherently calculation engines, and that professionals must work within their competence, citing the competence duties in Part 2A regulations 11F to 11H.
Rooftop chillers, condensers, air handling units, heat pumps and photovoltaic arrays are usually selected and positioned by the MEP designer. Plant weights, operating loads, vibration and support arrangements come from MEP information. Retrofit and decarbonisation work can mean more plant being added to existing roofs, often by teams without a structural engineer appointed at the outset. The temptation is to ask an AI tool whether the roof will take it. That is exactly the shortcut the report warns against.
Early feasibility work is where this risk is highest. A client may ask whether a building can take air source heat pumps or a new chiller before any structural engineer is appointed, and a quick AI answer can shape the whole scheme. Answer feasibility questions by stating that structural capacity is unconfirmed and recommending a structural appointment.
Illustrative scenario: on a fit-out, an MEP consultant replaces a condenser with a heavier heat pump unit and uses an AI chatbot to confirm the existing roof slab is adequate, without telling the landlord's structural engineer. Cracking appears in the ceiling of the flat below after installation. The plant has to be removed and relocated on a new steel frame, the residents are disrupted, and the landlord and tenant both claim against the consultant for investigation, remedial works and delay.
The interface failure is as important as the AI error: nobody with structural competence was asked.
A claim arising from plant information given negligently, such as wrong weights or locations, usually falls within an MEP engineer's PI, subject to the wording. A claim arising from structural advice an MEP firm was not competent or appointed to give may be more difficult, because policies are written around your declared professional activities and insurers may question work outside them. See how M&E and structural PI differ and the peer review failure scenario.
Related: AI-written specifications, AI heat pump design and LLMs for structural calculations. Back to AI in building services engineering.
Apex arranges PI for building services engineers. A broker reviews your cover and how your policy treats work at the interface with structural design.
No. The report describes an architect relying on AI-generated calculations to justify heavy plant on an existing roof. The lessons apply to MEP designers because plant weights and positions usually come from building services design.
You should not rely on it. The CROSS panel said large language models are not inherently calculation engines, and structural capacity needs a competent structural engineer. Give them accurate plant data instead.
Plant weights, operating loads, footprints, support points, vibration data and any changes from substitutions, with the source of each figure. This helps the structural engineer assess capacity properly.
It depends on the wording and your declared activities. Claims from MEP design are usually covered subject to the terms, but advice outside your discipline may raise questions with insurers.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.