Commercial AI tools are already on most desks. The risk isn't the tool. It's what gets typed into it. Every input is a disclosure decision, and it's made before you open the session, not after.

Before you open a tool

Use only the tools your organization has approved, under its written AI policy. If there's no policy or approved list, ask before you type anything that isn't public.

Sort every input into one of three tiers

The three tiers of procurement data
Tier What to do Examples
SAFE May go into an approved tool. Information that is already lawfully public: a posted solicitation, your published procurement code and policies, vendors’ public websites and capability statements. Inputting it discloses nothing new.
CAUTION Remove the identifying or sensitive specifics, then use only an approved tool. Internal drafts and notes that contain names, prices not yet public, or parties’ positions.
NEVER Stays out of every AI tool unless your agency has authorized that tool, in writing, for that kind of information. Information your procurement code, public records law, or other law makes confidential or exempt from disclosure. That can include bids and proposals during evaluation, vendor trade secrets and proprietary material, and personal information (PII). Anything your agency has marked confidential.

Which information is confidential, and for how long, is set by your state’s procurement code and public records law, and sometimes by local rules. When you’re not sure, treat it as NEVER and ask your counsel or records officer.

Your prompts may be public records

Public records laws generally reach records made or received in the course of public business, whatever their format. What you type into an AI tool, and what it gives back, may be one of them. Many tools keep a history of every session. Before you use a tool for agency work, find out whether your agency treats that history as a record, how long it has to be kept, and who answers a request for it. Then write every prompt as if it could be released.

How to anonymize a CAUTION input

1.Replace the names of people, companies, and agencies with roles: “the vendor,” “the department.”
2.Remove or round dollar values, dates, and locations.
3.Strip details that identify the purchase when read together, even if each one looks harmless.
4.Read it as an outsider. If you could still tell who or what it is, it isn’t anonymized yet.

One prohibited practice, whatever the tier

A vendor document can carry instructions hidden in white text, alt text, or metadata that change what an AI tool tells you about it. You can’t see them, and the tool may follow them. And you can’t audit that after the fact. So a vendor-submitted document never goes into any AI tool, including an enterprise tool your agency has approved, for any purpose, even anonymized. The workaround is to use AI earlier: build your evaluation criteria and their rationale before proposals arrive, then evaluate the proposals yourself.

Write it down

For any AI-assisted work that touches a purchase, record which tool you used, what went in, and what you kept or changed. Keep that record with the procurement file. The file has to stand on its own if someone reviews it later, or asks for it.

AI assists. Humans decide.

This primer is a working guide, not legal advice. Your organization’s AI policy, your procurement code, your public records law, and your counsel govern.

Keep going

Get the free AI in Procurement: 2026 Readiness Brief →

Where AI can help assemble procurement work, and where judgment stays with you.

Commercial GenAI for Public Sector Procurement

Five self-paced modules for federal, state, and local procurement professionals. Each module is $395 and ends with a certificate of completion.

Module 1: Using GenAI Safely in Government Procurement. Available now.

Covers data classification, anonymization, and documentation in depth, for federal, state, and local buyers.

Module 2: Regulatory Compliance Across Government Levels. Available now.

Federal acquisition under the Revolutionary FAR Overhaul, state procurement codes, local requirements, ethics, public records, and records management.

Module 3: Prompt Engineering for Government Procurement. Available now.

Prompt fundamentals, procurement-specific prompt patterns, and evaluating and improving outputs.

See the full course →