Using AI in professional advice: responsibility still needs a human owner

Automation does not sign the mandate

When an architect, consultant or engineer uses generative AI, the client still contracts with the professional business. An inaccurate calculation, omitted constraint or copied material can produce financial loss, delay or intellectual-property claims. Professional liability responds only to covered legal liability and policy terms; it does not certify the output or fund ordinary rework. The company should identify where AI is used, what data enters the system and which qualified person approves the result. Calling the tool a “drafting assistant” does not remove responsibility if staff rely on it without meaningful review.

Control data and validation

Classify client information before entering any external service and check contracts on retention, training use, location and subcontractors. Prohibit confidential or personal data in unapproved tools. For each high-impact output, retain prompt context, sources, version and human validation. Use independent calculations for safety-critical or regulated work. Tell clients about material tool use when contract, profession or transparency requires it, without overselling capabilities. Update engagement letters to describe deliverables, assumptions and reliance. A reproducible review trail helps distinguish a professional judgment from an unexamined machine suggestion.

A hypothetical design error

A small architecture firm uses AI to summarise local constraints for an early concept. The summary omits a setback rule, and the team copies it into a client presentation without checking the official plan. Redesign causes delay and extra fees. The firm notifies its professional-liability insurer, preserves the tool record and examines whether the client’s claim is covered. It also separates insured third-party loss from its own cost of correcting work. Future projects require a named architect to verify every regulatory statement against primary sources. The example does not decide negligence or coverage.

Speed competes with assurance

AI can reduce routine drafting time and reveal alternatives, but verification can consume much of that saving for consequential advice. Blocking every tool limits productivity; uncontrolled adoption creates confidentiality and accuracy risk. A specialist AI endorsement may address some exposures, while cyber and professional-liability policies cover different harms and exclusions. Contracting with a vendor transfers only responsibilities stated and enforceable in that agreement. The business should price review time into the service and reject uses where it cannot independently test the result.

Create an approved-use register

List tools, owners, permitted data, use cases, review standard and renewal date. Map each use to client contracts, professional rules, data protection, cyber controls and insurance notifications. Ask the liability insurer how AI-assisted work, intellectual property, confidentiality and pure financial loss are treated. Train staff with examples of unacceptable reliance and make reporting near misses safe. Audit a sample of outputs quarterly. AI may improve service, but accountability must remain visible from source to sign-off; no insurance policy makes an unverified answer professionally sound.

The firm should also test how a client complaint travels from the project team to management, counsel, the insurer and any technology supplier. Keeping the original prompt, source documents, revisions and final approval can distinguish a controlled professional process from an unexplained automated output when facts are later disputed.