Aipedia: Structural and civil engineers

AI design checking and peer review

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

AI can flag issues in a design, but it cannot replace an independent check by a competent engineer, and IStructE expects a senior engineer to review AI-produced design. A check that leans on software risks automation bias and gives clients false assurance.

Part of: AI in structural and civil engineering

In short

Design checking exists to catch what the designer missed. When AI does the checking, two things happen: the software misses errors outside its training or rules, and the human reviewer tends to trust its green ticks and look less hard. That is automation bias. IStructE expects senior review of AI-produced design and regular evaluation of tools against current standards. If a reviewed design later fails, claims usually focus on whether the check was carried out with reasonable skill and care, and PI typically treats a peer-review failure as a professional negligence claim, subject to the wording.

How engineers use AI as a checker

Practices are using AI in several checking roles: code-checking modules that flag overstressed members, model QA tools that look for missing loads or supports, LLMs asked to "review this calculation for errors", and drawing checkers that compare general arrangement drawings with the analysis model. Some firms use these to support internal peer review; others are tempted to let them stand in for it on small jobs where fees are tight.

The attraction is speed and coverage: software can test every member in a large model, which no human checker will do by hand. The limit is that it tests only what it was built or prompted to test. It has no view on whether the structural concept is sensible, whether the load path is complete, or whether the model represents what will actually be built.

The duties that apply

IStructE AI guidance: a senior engineer should review any AI-produced structural design, tools should be evaluated against current standards, and firms should keep an AI risk matrix. IStructE guidance.
IStructE Code of Conduct, Article 6: appropriate skill and judgement, including a quality management system. A checking procedure is part of that system. Code of Conduct.
Part 2A reg 11K: a designer of part of a project must consider how their design relates to other design work and report compliance concerns.
CROSS-UK: all AI outputs need competent checking, review, validation and independent assessment. CROSS Report 1554.

What can go wrong

Automation bias is the main failure mode. A checker who sees no warnings tends to sign off faster, while the software may not test the thing that matters: load paths, robustness, stability during construction or a connection detail outside its scope. An AI asked to review its own output may confirm its own mistake. CROSS 1554 showed errors persisting across several revisions of an AI-written report because nobody competent was checking.

Illustrative scenario: a firm adopts an AI model-checker and reduces manual peer review to a sign-off sheet. The tool checks member utilisation but not lateral stability during a transfer stage. A partial collapse of temporary propping follows, and the client, contractor and the firm's own reviewer all ask why the check did not identify it.

How PI usually responds

A claim that a check was inadequate is a claim of professional negligence, and PI commonly responds on that basis. The test is usually the Bolam standard: would a responsible body of engineers have checked the design in that way? Relying solely on software that did not test the relevant issue may be hard to defend. The policy wording, exclusions and the accuracy of what you told insurers about your QA procedures all matter. Our peer-review failure scenario walks through a typical claim, and engineer negligence claims examples gives wider context.

Practical controls

Treat AI checking as an extra layer, never as the independent check.
Define in writing what each tool checks and, more importantly, what it does not.
Require the human checker to rework critical elements without looking at the AI result first.
Record senior review on every AI-produced design, as IStructE recommends.
Review the tool against current standards at set intervals and log the result.
Give staff a route to report a tool that seems unreliable and authority to stop using it.

How Apex helps

Apex arranges PI for structural and civil engineers. A broker reviews your cover and how your policy and proposal form describe AI-supported checking procedures.

Sources

Frequently asked

Can AI count as an independent design check?

Not on its own. IStructE expects a senior engineer to review AI-produced design and CROSS-UK says all AI outputs need competent checking and independent assessment. AI can support a checker, but the judgement and sign-off should come from a competent person.

Should we tell clients that AI was used in checking?

IStructE says firms should tell clients what AI is used for and share their AI policy with clients, architects, MEP engineers and Building Control. Explaining that AI supports, but does not replace, the independent engineering check also avoids overstating the assurance a client is getting.

What is automation bias in design checking?

It is the tendency to trust an automated result and scrutinise less. A checker who sees a clean software report may skip the manual review that would have caught a load path or stability error outside the tool's scope.

Would PI cover a claim that our AI-assisted check was negligent?

Usually PI responds to claims that professional work, including checking, fell below a reasonable standard. Cover depends on the wording and on accurate disclosure of your QA procedures, so describe your use of checking tools honestly at renewal.

Check your PI covers how you use AI

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 callback

Apex 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.