An AI policy should give employees practical answers before they use a tool on a live project. It should identify which systems the firm has approved, which tasks are allowed, what information can be entered, and who is accountable for reviewing the output. A short policy works best when employees can apply it to an actual proposal, calculation, report, or client email.
Start by listing permitted uses, such as organizing nonconfidential notes or drafting an internal outline. Define restricted uses and require approval before a new tool or workflow handles project material. Prohibit staff from entering client-confidential, personal, controlled, or sensitive information unless the firm has confirmed that the client permits it and the tool’s safeguards meet the firm’s requirements. Explain where staff can find the current approved-tool list and how to report an accidental disclosure.
For technical and client-facing work, require a qualified person to check facts, calculations, citations, assumptions, and tone against original sources. State clearly that an AI system cannot approve or seal engineering work, accept responsible charge, or replace a licensee’s judgment. Set recordkeeping rules for prompts, source documents, review comments, and final approvals when those records matter to project quality or a contract.
Owners should assign policy responsibilities instead of leaving them vague. For example, a technical principal can own engineering review rules, while a security lead tracks tool terms and access. Ask counsel to review contract, privacy, intellectual property, and disclosure requirements. Ask project managers what examples staff need to recognize common problems, such as fabricated citations or stale code references. Train staff with those examples and revisit the policy when tools, client requirements, or applicable rules change.
Contracts, applicable law, licensed PE judgment, and local codes govern.
Have the owners state which information staff may enter into AI tools, which uses need review, and who is accountable for the final work product. Cover client confidentiality, project records, copyright, retention of prompts and outputs, and approval for tools that process firm data. For example, staff might use an approved tool to organize public specification text but need manager approval before submitting client plans or calculations. State that AI output cannot replace licensed PE judgment, code review, or required checking, and provide a process for reporting an accidental disclosure.
This is general education, not legal or engineering advice.
