StrategicalAI.org

The Uniform Guidance (2 CFR 200) wasn’t written for AI tools. That doesn’t mean AI tools are unallowable under federal awards — it means the existing cost principles apply, and you have to do the analysis. Most organizations skip the analysis and hope. Hope is not a compliance strategy.

Here are five questions to answer in writing before you charge any AI tool cost to a federal award. The answers should live in a memo, dated before the first charge.

Question 1: Is this cost necessary for the program?

Necessary, not merely useful. Under 2 CFR 200.403, an allowable cost must be reasonable for the performance of the award. “Reasonable” means a prudent person, under the circumstances, would have incurred this cost.

Write down: what specific program activity does this tool support? What was the prior cost of doing this activity without the tool? If you can’t articulate the activity in plain language and connect it to the program’s approved scope, you don’t have a necessary cost — you have an aspirational one.

Question 2: How is the cost allocated if it serves multiple awards?

2 CFR 200.405 governs allocability. If a single AI tool serves two or more awards (e.g., a Microsoft 365 Copilot subscription used across multiple federally funded programs), the cost has to be allocated using a defensible methodology.

Common methodologies: FTE-time on each award (“50% of subscription cost is charged to Award A because 50% of the staff using the tool work on Award A”), output-based (“40% of grant proposals supported by the tool went to Award B”), or transaction-count.

Document the methodology BEFORE you make the first charge. Auditors look for the time-stamped memo establishing how you decided to split the cost. A methodology decided in retrospect is a red flag.

Question 3: Is the cost direct or indirect?

2 CFR 200.413 and 200.414 govern the direct/indirect classification. The choice has cash-flow implications and audit implications.

A program-specific AI tool (e.g., an AI subscription used only by case managers on Award X) is typically a direct cost. An organization-wide AI tool (e.g., Microsoft 365 Copilot used across all administrative functions) is typically indirect and rolls into your indirect rate.

The consistency rule (2 CFR 200.403(d)) is the trap: whatever classification you choose, treat the same cost type the same way across all awards. Inconsistent treatment of the same line item is the audit finding you most want to avoid.

Question 4: How will the AI use be disclosed in performance reporting?

2 CFR 200.328 and 200.329 govern performance reporting. The principle is honest representation of program outputs.

If AI materially affected how an output was produced — an AI-summarized participant interaction, an AI-generated grant narrative, an AI-recommended case action that staff acted on — the performance report should be transparent about that role. Not in a confessional way; in a precise way. “Outputs include AI-assisted drafting reviewed by program staff before submission” is honest. “Outputs produced by the program team” without disclosure of AI assistance is, at best, misleading.

The line you must not cross: representing AI-generated content as human-authored in funder evaluations. That’s a misrepresentation that becomes a fraud question fast.

Question 5: What documentation will an auditor see?

Imagine your single-audit firm asking, three years from now, “explain this AI tool subscription line item.” What do you hand them?

Minimum documentation: (1) the written AI Use Policy adopted by your board; (2) the approved-tool list showing this tool, the tier, the data class it’s approved for, and the date of approval; (3) the cost-allocation methodology memo; (4) the AI-use log if the tool is used at the Restricted tier; (5) any grants-officer correspondence approving AI scope or costs.

The Federal Grants & AI Compliance Quick Reference includes a checklist version of this. Print it out, fill it in, and keep the paper trail in the file you’d hand to the auditor.

Program-specific stricter rules apply

Many federal programs have rules that are stricter than the general Uniform Guidance baseline. The stricter requirement always governs. A non-exhaustive list of where AI use gets more constrained:

• HIPAA-covered programs: AI tools handling PHI need a Business Associate Agreement. Most consumer AI tools cannot offer one.
• FERPA-covered programs: educational-record AI use is constrained; verify before integration.
• Head Start: child-data protections are stricter than the general PII baseline.
• WIOA-funded programs: state PIRL integrations have separate data-sharing terms.

When in doubt, email your grants officer

The most undervalued compliance tool in federal grant management is a short, specific written question to your grants officer, with their written response on file. A 200-word email exchange that confirms a specific AI cost is allowable under your award is worth more at audit time than 20 pages of internal memos.

Most grants officers will answer specific questions specifically. Take advantage of that.

Need a compliance review for AI integration in your specific federal award? Visit strategicalai.net.