Understanding the Core Questions
Federal procurement teams are grappling with how to integrate AI into their processes while staying compliant with the Federal Acquisition Regulation (FAR). The Court of Federal Claims' decision in the TRAX International case highlighted the need to include AI-generated evaluations in administrative records. This has raised questions about what's permissible, what must be documented, and who ultimately makes decisions when AI is involved.
The FAR Council's recent proposed rules don't specifically address AI in bid evaluations, leading to operational confusion. Let's explore the key questions your team needs to address.
Q1: Do You Need to Disclose AI Use to Bidders?
Yes, you must inform bidders if you're using AI in evaluations and document the tools and methods before starting. The Court of Federal Claims has ruled that AI-generated evaluations are part of the administrative record, making them discoverable in a protest. Disclosing AI use upfront helps prevent claims of undisclosed evaluation criteria.
Document the AI tool's name and version, the evaluation factors it will analyze, and whether it's generating recommendations or just summarizing data. This documentation isn't about asking permission; it's about creating a defensible record showing you controlled the tool's outputs.
Q2: Can a Contracting Officer Rely Solely on AI Recommendations?
No, a government official must independently verify and decide on every AI-generated recommendation. AI can inform decisions but cannot make them. The contracting officer should review AI outputs, apply professional judgment, and document their conclusions. The administrative record must show human judgment at every decision point.
Q3: How to Handle AI Errors or Hallucinations?
Document any AI errors or hallucinations, explain why they were rejected, and keep them in the contract file. The TRAX case showed that all AI outputs are part of the administrative record, even if deemed unreliable. Note any errors in the file and clarify that the evaluation was based on actual sources, not flawed AI outputs.
Q4: Verifying AI Recommendations in High-Volume Evaluations
Focus on verifying AI outputs that influence decisions, not every intermediate calculation. Integrate verification into your workflow: AI generates output, evaluators review source material, confirm or correct AI output, and document the decision. For high-volume tasks, consider sampling to verify a representative subset of AI-flagged issues.
Q5: What Must Be Included in the Contract File?
Include every AI-generated output related to the evaluation, documentation of the tools used, and the disposition of each AI recommendation. The contract file should detail the AI tool identification, evaluation factors, AI-generated summaries or recommendations, and the contracting officer's independent determination for each output.
Q6: Should You Avoid AI Until FAR Provides Guidance?
No, establish internal governance standards now. The absence of specific FAR provisions doesn't prohibit AI use; it means applying existing procurement principles to AI tools. Decisions must be made by authorized officials, evaluations documented, and the administrative record must support the award decision.
Develop your AI governance framework around these principles. Document your framework and apply it consistently. This preparation will provide a defensible record and operational experience when FAR guidance arrives.
Where to Go for More
The FAR Council's broader Federal Acquisition Regulation overhaul is ongoing. Track FAR case releases and proposed rules through the Federal Register.
For AI governance frameworks, review NIST AI RMF's guidance on transparency and accountability in AI-assisted decision-making. ISO/IEC 42001's requirements for AI Management Systems include controls for documenting AI use and maintaining human oversight.
Focus on maintaining a clear documentation trail: what the AI did, what the human decided, and how the decision was verified. This record is crucial for surviving a protest.
