FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: IT consultants and developers
AI-generated specifications, documentation and test suites can look complete while missing the requirements or edge cases that matter, letting defects through acceptance. Because these documents define what was agreed and tested, gaps in them shape who wins a later dispute.
Part of: AI in IT consultants and developers
In short
Teams now use AI to draft requirements from workshop notes, generate technical documentation and write test suites. The risk is plausible but incomplete output: a requirement silently dropped, documentation describing behaviour the system does not have, or tests that check the code does what it does rather than what it should. BCS clause 3.5 prohibits misrepresenting performance information, and 3.1 requires due care. Disputes often centre on acceptance testing. Technology PI usually responds to claims arising from negligent specifications or testing, subject to the wording.
Business analysts paste workshop transcripts into a generative AI tool and ask for user stories. Developers generate API documentation and user guides from code. Test engineers ask AI to produce unit, integration and acceptance tests from requirements or directly from the code. Each step saves time. Each also adds a layer where the AI may summarise, infer or invent, and the output is often reviewed less carefully than code because it reads well.
The BCS Code of Conduct says members must not misrepresent performance information (3.5). Documentation claiming capabilities the system lacks, or test reports suggesting coverage that was not achieved, could fall foul of that. Clause 3.1 requires due care and diligence, and 3.3 requires you to accept responsibility for your own and supervised work, including AI-assisted outputs a junior analyst produced.
The AI Cyber Security Code of Practice expects documentation of data, models and prompts (principle 8) and appropriate testing and evaluation (principle 9) where you build AI systems. The Software Security Code of Practice applies to software vendors generally, so secure development expectations do not fall away because test cases were machine-written.
Illustrative scenario: a consultancy delivering a case management system uses AI to turn workshop notes into requirements and then to generate acceptance tests from those requirements. A requirement for an audit trail on deleted records is lost in the summary. The tests, built from the same requirements, never check for it. The client signs off. A year later a regulator asks the client for deletion records it cannot produce, and the client claims against the consultancy for the cost of remediation.
Another pattern is self-confirming tests: AI-generated tests built from existing code pass because they encode the bug as expected behaviour.
Acceptance is often where these issues meet the contract. If the client signed off against agreed tests, the consultancy may argue the client accepted the system. The client may respond that it relied on the consultancy's expertise to write adequate tests, or that documentation misdescribed what was delivered. Who wrote the acceptance criteria, who reviewed them and what the contract says about latent defects will matter. See software delivery failure claims and scope of duty in IT claims.
Technology PI is commonly written to cover negligent acts, errors and omissions in your services, which includes analysis, documentation and testing as well as coding. Whether AI drafted the document is not usually relevant to cover, subject to the wording. Claims based on express warranties, for example that a system "meets all requirements", may fall partly outside cover if they go beyond your legal liability. Accurate descriptions of your services on the proposal form help.
Some clients now want to know where AI was used in deliverables. Under PPN 017, public sector buyers may ask suppliers to declare AI used in bids and in service delivery, and may verify AI-assisted content. A specification or test plan that was largely AI-drafted and not declared, where a declaration was requested, could create a dispute separate from any defect. Keeping a simple record of which deliverables were AI-assisted, and how they were reviewed, makes those questions easy to answer and supports the honesty expected by BCS clause 3.5.
Apex arranges technology PI for IT consultancies and software teams. A broker reviews your cover and how your policy treats claims over specifications, documentation and testing. See the IT consultants AI hub and AI-assisted data migration.
Only with human review. AI-generated tests often mirror the code or the requirements they were given, so they can miss requirements that were dropped earlier. Include some tests written independently from the client's real business needs.
Clause 3.5 prohibits misrepresenting performance information. Documentation or test reports that describe capabilities or coverage the system does not have could breach it, whether AI or a person drafted them.
Sign-off helps but is not always conclusive. A client may argue it relied on your expertise to design adequate tests, or that defects were latent. The contract terms on acceptance and latent defects will be important.
Check the contract. Public sector buyers using PPN 017 may ask suppliers to declare AI use in service delivery. Even where not required, telling the client helps it review AI-drafted requirements carefully and reduces later arguments about what was agreed.
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.
