Artificial Intelligence Compliance and Risk Management Plan.docx

Artificial Intelligence (AI) Use Compliance and Risk Management Plan Purpose: Use this when a contractor plans to use, may use, or later proposes to use AI in contract performance, but the primary deliverable is not the development or delivery of an AI system.

1. Background Name of AI tool/system:

Contractor using the AI:

Subcontractor or service provider, if any:

Contract task/function supported:

Describe how AI will be used in contract performance:

Is AI use required to perform the work, or optional?

Required Optional Unknown AI classification:

Generative AI Agentic AI Classical/Predictive Machine Learning Computer Vision Natural Language Processing Reinforcement Learning Other: [Insert] Is this use limited to contractor internal operations?

Yes No

Will the AI touch Government information, systems, identities, networks, or operations?

Yes No Unknown Was the AI system procured to support this particular contract per requirements in the Statement of Work?

2. Who owns and supports this use?

Contractor AI owner:

Contractor technical support contact:

Contractor privacy/security contact:

CDC/HHS program office using or receiving outputs:

Expected users:

Contractor staff only CDC/HHS staff Public users Beneficiaries, patients, applicants, regulated entities, or other external parties Other: [Insert]

3. Tool background Tool or product name:

[Insert] Developer/vendor:

[Insert] Is the AI technology open-source, proprietary, Government-furnished, or other?

☐ Open-source ☐ Proprietary/commercial ☐ Government-furnished ☐ Contractor-developed ☐ Subcontractor-provided ☐ Other: [Insert] If the system uses an LLM or foundation model, what is the primary model?

[Insert] Who provides the LLM/foundation model?

[Insert] What model options are available?

[Insert] Does it require an API key?

☐ Yes ☐ No ☐ Unknown Attach or link, if available:

☐ Model card ☐ System card ☐ Data card ☐ Vendor security documentation ☐ Evaluation summary ☐ Terms of service ☐ Privacy/data retention policy ☐ Audit-log documentation

4. Where is it hosted and authorized?

Hosting environment:

☐ Contractor environment ☐ Commercial SaaS ☐ FedRAMP-authorized cloud ☐ CDC/HHS environment ☐ Local/offline environment ☐ Other: [Insert] Where is the system hosted?

[Insert country/region/cloud/provider] Is the product directly FedRAMP authorized or available through a FedRAMP-authorized instance?

☐ Yes ☐ No ☐ Not applicable ☐ Unknown If no, is it undergoing FedRAMP authorization?

☐ Yes ☐ No ☐ Unknown

5. What data will be used?

Will the AI process, store, transmit, summarize, analyze, or generate Government information?

Yes No Unknown Types of data involved:

Public information Government nonpublic information

PII

PHI

Procurement-sensitive information Proprietary/business confidential information ☐ Other: [Insert] What information will be entered into the AI tool?

[Insert] What outputs will the AI generate?

[Insert] Will prompts, uploads, outputs, logs, or user interactions be stored?

☐ Yes ☐ No ☐ Unknown Retention period:

[Insert] Will Government data, prompts, outputs, or logs be used to train, fine-tune, improve, or benchmark any model?

Yes No Unknown Will any data, prompts, outputs, or logs be shared with other users, tenants, vendors, subcontractors, or third parties?

Yes No Unknown Explain storage, use, retention, deletion, and sharing:

[Insert]

6. Is this high-impact?

Could the AI output materially affect an individual’s or organization’s rights, benefits, services, health, safety, access, eligibility, enforcement status, or other significant interests?

Yes No Unknown Could CDC/HHS personnel rely on the AI output as a principal basis for a decision or action?

Yes No Unknown Is the AI used only for internal drafting, summarization, or administrative support with human review before use?

Yes No High-impact rationale:

[Insert]

7. What safeguards are in place?

How will contractor personnel verify AI outputs before relying on them?

What human review is required before AI-generated content is submitted to CDC/HHS?

How will the contractor prevent unauthorized disclosure of Government information?

How will the contractor prevent misuse, unauthorized use, or corruption of the AI system?

How will the contractor protect privacy, civil rights, and civil liberties?

What internal policies govern this AI use?

8. Monitoring and misuse protection What is monitored?

Accuracy/performance Performance degradation or drift Data leakage signals Abuse/misuse Prompt injection Unauthorized access Other: [Insert] Does CDC/HHS have access to monitoring information, audit logs, or summary reports?

Yes No Upon request Not applicable How are issues detected, escalated, and corrected?

When will CDC/HHS be informed?

What is the standard operating procedure for incidents or unacceptable outputs?

9. Required contractor affirmation The contractor affirms that:

AI will not be used unless approved by the Contracting Officer.

The contractor has disclosed all planned or potential AI use.

The contractor will not enter Government information into unapproved AI tools.

The contractor will not use Government data to train, fine-tune, or improve an AI model unless expressly authorized.

The contractor will maintain and update this AI Use Card throughout contract performance.

The contractor will notify the Contracting Officer and COR before changed, expanded, or new AI use.

The contractor will prevent unlawful discriminatory use or outputs.

The contractor will comply with applicable privacy, security, data rights, records, civil rights, and AI requirements.