Key takeaways
- California and federal trade secret law protect information only when the owner makes reasonable efforts to keep it secret.
- Fee proposals, rate tables, unit costs and past bids should never go into an unapproved AI tool.
- AI proposal writing is safest when the tool reads only a clean proposal library with pricing removed.
- Treat public AI services as publicly accessible systems under NIST SP 800-171 requirement 3.1.22 and keep CUI out of them.
- Web filtering, DLP, sensitivity labels and new-app alerts enforce the proposal AI rule.
Is AI proposal writing safe for AEC firms?
Yes, AI proposal writing is safe for AEC firms when the work runs in an AI tool the firm controls and pricing stays out of everything else. The risk sits in the data. A proposal team that pastes fee schedules, unit rates, labor multipliers or a losing bid into a personal AI account has moved some of the firm's most valuable information to a vendor the firm has no contract with.
Proposal writing is one of the most natural uses of AI at a design or construction firm. Teams rewrite project descriptions, tailor resumes, summarize RFPs, draft RFP responses and write cover letters under deadline. The files that make AI useful for proposals, past submissions and pricing, are the same ones a firm least wants to leave its control.
Why do pricing and past bids need special handling?
Pricing and past bids need special handling because their legal protection depends on the firm keeping them secret. Under California's Uniform Trade Secrets Act, Civil Code section 3426.1(d) defines a trade secret as information, including a compilation, that derives independent economic value from not being generally known to the public or to others who could obtain economic value from it, and that is the subject of efforts that are reasonable under the circumstances to maintain its secrecy. The federal Defend Trade Secrets Act uses a similar test: 18 U.S.C. 1839(3) requires that the owner has taken reasonable measures to keep the information secret.
Whether a particular firm's rates qualify is a question for counsel. What the statutes make clear is that the firm's own conduct is part of the test. A firm that lets staff paste its rate tables into any AI tool they like will have a harder time showing reasonable efforts than a firm with a written rule and a control that enforces it.
What proposal content should never go into an unapproved AI tool?
Draw the line around content that would help a competitor or harm a client if it left the firm. For most AEC firms, that list includes:
- Fee proposals, rate tables, labor multipliers and markups.
- Unit costs, estimates and subcontractor quotes.
- Past bids and proposals, won or lost, with their pricing.
- Client names, contacts and project details marked confidential in an RFP or contract.
- Security, utility or facility drawings, and any drawings from defense projects.
The last item has its own rules for firms on defense work. NIST SP 800-171 Rev. 2 requirement 3.1.22, the revision CMMC Level 2 assessments use, requires firms to control Controlled Unclassified Information posted or processed on publicly accessible systems. Treat a public AI service as one of those systems and keep CUI out of it. Our guide to keeping CUI out of AI tools on defense projects covers the approved options.
Which AI proposal tools are safe to use?
A safe AI proposal tool is one the firm has approved, on firm accounts, under business terms that bar the vendor from training on firm data. Consumer accounts on free AI services rarely meet that test, because the account belongs to the individual and the firm cannot see, export or delete what was shared.
Business plans usually offer better terms. Microsoft, for example, states that with enterprise data protection in Microsoft 365 Copilot and Copilot Chat, prompts, responses and data accessed through Microsoft Graph are not used to train foundation models (Microsoft documentation, checked October 2026). Read the terms for any tool the firm approves and record what you found: whether the vendor trains on firm data, how long it keeps prompts and files, who at the firm can see them, and how the firm gets its data back or deleted.
Approval should also cover where the source material lives. If the tool can search the firm's past proposals, limit that access to the proposal library and the people who need it. A tool that can read every file a user can read will happily surface last year's fee schedule in a draft for a different client. The same approach applies to AI design and rendering tools that touch client drawings.
How do you keep pricing out of AI drafts and RFP responses?
Keep pricing out of AI drafts by separating the narrative from the numbers. Most of a proposal or RFP response is narrative: firm history, project approach, team resumes, relevant experience. That content can go through an approved AI tool with little risk once client-confidential details are removed. Pricing is a short section that changes with every pursuit and is better built outside AI by the people accountable for it.
A simple workflow keeps the two apart:
- Keep a clean proposal library of approved narrative, resumes and project descriptions with pricing and confidential client details removed. Point the AI tool at that library only.
- Build fee and pricing sections in the firm's estimating or accounting system, and add them to the final document after the AI drafting is done.
- Have the proposal lead check every AI-drafted section against the RFP and the source projects before submission. AI tools can invent project details, square footage or awards that sound right and are wrong.
- Store final submissions, including pricing, in a restricted folder with access limited to the pursuit team and leadership.
Which controls enforce an AI proposal rule?
Web filtering, data loss prevention, sensitivity labels and access controls enforce the rule, because written rules work better with technical controls behind them. The CIS Controls v8.1 safeguards for software inventory, data protection and web filtering map directly to this problem:
- Web filtering blocks or warns on unapproved AI sites from firm devices.
- Data loss prevention detects rate tables, estimate files and labeled documents being pasted or uploaded to AI sites. Microsoft Purview, for example, can block text and file uploads to unmanaged AI apps through Edge for Business on managed Windows devices.
- Sensitivity labels mark pricing files and past bids so DLP rules can find them.
- New-app alerts flag when someone signs up for an AI service with a firm email address.
- Zero trust access limits who can reach the proposal library and the pricing folder, including from home or a jobsite. Here is how to replace your VPN with zero trust network access.
What should an AEC firm do this quarter?
Start with five steps that put most firms in a defensible position before the next round of pursuits.
- Write a one-paragraph proposal rule: approved AI tool, firm accounts only, and the list of content that never goes into AI. It can sit inside a broader AI policy for design, proposals and field teams.
- Approve one AI tool for proposal work and record its data terms.
- Build or clean the proposal library so it holds narrative only, and restrict the pricing folder.
- Label pricing and past-bid files, then turn on DLP rules for those labels.
- Brief the marketing and pursuit team, and add an AI review step to the proposal checklist.
How s90 helps
s90 helps architecture, engineering and construction firms use AI on proposals without handing pricing to outside services. We set up approved AI tools on firm accounts, data loss prevention rules for rate tables and past bids, alerts for new AI sign-ups, and zero trust access to proposal and pricing folders. If you want a quick review of how your proposal team uses AI today, we are happy to walk through it.
