CIBSE warns engineers of PI and legal risks with AI

CIBSE’s wide-ranging and detailed position statement runs to 42 pages. The institution notes: “It would be counter-productive to curtail or discourage the adoption of new ways of working, but we also have an obligation to remind engineers of the legal and professional risks involved. This is the framing and approach that CIBSE’s AI working group took in developing this position statement – opportunity first, with a grounding in current and anticipated regulatory landscapes.”

The statement is explicit about the need for human oversight and the level of detail that should be applied when validating AI outputs: “All AI-supported workflows must therefore include explicit human oversight, documented quality assurance and sign-off by competent engineers, consistent with the human-oversight principles of the EU AI Act (EU, 2024). Accountability for engineering decisions cannot be delegated to AI systems.

It continues: “Statutory and professional roles such as Authorised Persons, Authorising Engineers, Competent Persons, Responsible Persons under fire safety legislation and duty holders under the Building Safety Act 2022 can only be held by people; AI may assemble, organise and interrogate the evidence on which those roles rely, but it cannot be the assurer. For accountable practices across the AI lifecycle, organisations should ensure they have a clear chain of human responsibility, auditability and answerability for AI-assisted engineering work. All parties involved in projects using AI should be aware of their legal duties and make certain they comply with them.

“The level of assurance should increase with the consequence of error. For safety-, compliance-, comfort-, energy- or carbon-critical decisions, validation should use authoritative source material and, where practicable, independent calculation, simulation, measurement or deterministic checks, rather than relying on a second generative-AI response as verification.”

The statement reiterates established concerns about AI use and intellectual property, ownership of outputs and standard appointment terms, but its also tackles the PI issue. “The insurance industry is still trying to understand the full extent of the implications of AI usage. Moving forward, insurers will want to understand the extent to which their insureds are using AI and what guardrails they have in place in relation to its use,” the statement reads. “Insureds should check whether their AI uses fall within the scope of their PI and cyber-risks cover; disclose material changes where required; and obtain clarification from their brokers or insurers where they are using AI in a material way in particular in any client deliverables.”

AI use complicates PI cover in several ways, according to CIBSE:

CIBSE provides a warning that DC+ has not seen anywhere else: “Insurers are particularly sensitive to systemic risk – where a single failure causes multiple claims. AI creates exactly this scenario. We can imagine cases where a widely used AI design-checking tool has a flaw, a generative layout algorithm embeds a systematic error, or an optimisation model miscalculates performance assumptions. If dozens of firms use the same tool, claims may arise across multiple projects, and potentially across multiple insurers, from the same underlying cause. This kind of correlated loss is difficult to insure and may lead to higher premiums, policy exclusions and sub-limits for AI-related claims.”

Ahmed Hegazi, head of knowledge and digital at CIBSE, said: “We have always been at the forefront of building engineering, and our approach to AI is no different. We see enormous opportunities for AI to enhance engineering expertise, drive innovation and deliver better buildings. But in an industry where safety, competence and accountability are paramount, technology must support professional judgement, never replace it.

“Our new position statement, shaped by leading industry expertise and rigorous peer review, sets out the principles for responsible AI adoption. Importantly, this is just the beginning of the conversation. We want to work with our members and the wider industry to ensure AI develops as a force for engineering excellence.”

CIBSE will review its position on the responsible use of AI regularly.

CIBSE’s announcement follows a number of industry associations and institutions reminding members of their commitments when using AI. In the summer, the University of the Built Environment’s leadership forum, Built Environment Futures Assembly, said, “clients and procurement teams increasingly need to understand not only whether suppliers are using AI, but how they are using it, with what data, under what controls and with what assurance”.

Earlier this year, nima’s thinktank, the GIIG, warned the industry to be sceptical of AI’s benefits and wary of its risks. Last year, RICS became the first built environment institution to embed the ethical and correct use of AI into its standards.

Keep up to date with DC+: sign up for the midweek newsletter.