Issuance of HUD Acquisition Instruction 26‑02 – Acquisition of Artificial Intelligence and New Agency‑Specific Clauses

Other / Unclassified

This document, "AS-2604 Vendor Lock-In Protections," outlines specific requirements for contractors to ensure long-term agency control over systems, data, and Artificial Intelligence (AI) Systems, AI Services, and AI models. It mandates the implementation of safeguards to prevent vendor lock-in.

This document details critical contractual obligations related to AI systems and data. Key requirements for bidders include:

• Open Standards: Use of open standards and support for interoperability, with documentation required.
• Knowledge Transfer: Comprehensive documentation, training, and knowledge transfer to enable agency personnel or successors to manage, operate, and update AI systems.
• Data and Model Portability: Ensuring all created data, metadata, models, and configuration files are provided in a machine-readable, non-proprietary format for portability.
• Government Rights: Granting the Government sufficient rights for operational use, development, inspection, and reuse of custom-developed code, AI systems, models, and deliverables.
• Licensing Transparency: Full disclosure of all licensing terms, including third-party components, usage rights, renewal options, and costs.
• Pricing Transparency: Clear, itemized pricing for services, software, and ongoing support, including projected future costs.
• Transition Support: Providing support for a smooth transition to an alternative provider or internal capability upon contract expiration or termination.
• Subcontractor Compliance: Flowing down these vendor lock-in protection requirements to subcontractors.

Contractors must also provide an estimate of the cost to bring AI systems into compliance. Failure to adhere to these requirements may result in the Government exercising contractual rights, including stopping work until compliance is demonstrated.

Solicitation / RFP / RFQ

This document outlines "AS-2606 Requirements for High-Impact AI Systems," a clause intended for inclusion in solicitations and contracts above the micro-purchase threshold. It mandates that contractors comply with all applicable federal and agency risk mitigation requirements for any AI system or service designated as a High-Impact AI System. These requirements apply regardless of when the designation occurs.

Contractors must:

• Implement and document risk management and mitigation measures proportionate to the AI System's impact.
• Support the Government in pre-deployment testing and validation, providing access to system functionality, documentation, and evaluation results. Conduct their own pre-deployment testing.
• Implement continuous monitoring of AI System performance post-deployment, identifying adverse impacts on performance, security, privacy, civil rights, or civil liberties.
• Support the Government in developing and implementing appropriate mitigations and maintaining accurate system documentation.
• Support the Government in developing role-based training for users and operators on interpreting AI outputs, managing risks, and complying with policies.
• Ensure the AI System enables appropriate human oversight and review, including intervention, override, or suspension of automated outputs.
• Provide sufficient technical and non-technical documentation for Government oversight and audit, including purpose, benefits, cost information, data characteristics, potential impacts on privacy/civil rights/liberties, risk mitigation implementation, and reassessment schedules.
• Maintain logging and traceability of AI System inputs, outputs, and actions.
• Promptly report AI System incidents, failures, or identified harms and implement corrective actions.
• Notify the Government in advance of material updates, retraining events, or architectural changes that could alter the AI System's risk profile or performance.
• Flow down all High-Impact AI System requirements to subcontractors.

This document is a clause to be used in solicitations and contracts. Specific contract type, period of performance, and estimated value are not detailed within this clause itself.

Not applicable to this document, as it defines requirements rather than a submission process.

Not applicable to this document.

• A list of AI Definitions can be found on the AI@HUD page on hud.gov.
• Failure to adhere to these requirements could result in the Government exercising contract rights, including disincentives, liquidated damages, or actual damages. The Government reserves the right to stop work until compliance is demonstrated.

Other / Unclassified

This document is Acquisition Instruction (AI) 26-02, titled "ACQUISITION OF ARTIFICIAL INTELLIGENCE." It provides guidance for the procurement of safe and secure artificial intelligence (AI) to enhance government efficiency and mission effectiveness, ensuring AI models produce reliable outputs free from bias. The instruction aligns with various Executive Orders and OMB Memoranda related to AI.

This instruction is relevant to potential bidders as it outlines the U.S. Department of Housing and Urban Development's (HUD) policy and procedures for acquiring AI systems. Key aspects for bidders include:

• Policy Requirements: All AI procurements must adhere to the Federal Information Technology Acquisition Reform Act (FITARA) review process and involve an Integrated Project Team (IPT) to assess risks.
• Agency Specific Clauses: Solicitations and contracts will include specific clauses (AS-2602 through AS-2606) covering:
  • Privacy, Intellectual Property, and Government Use
  • AI Use and AI Model Transparency (requiring truth-seeking and ideological neutrality)
  • Vendor Lock-In Protections
  • Ongoing Testing and Monitoring
  • Requirements for High-Impact AI Systems
• Attachments: Bidders should be aware of Attachment 1 (AI Acquisition Best Practices) and Attachment 3 (AI Acquisition Process), which must be followed.
• Applicability: This instruction applies to HUD acquisitions above the micro-purchase threshold awarded on or after its issuance, and to options to renew or extend existing contracts.
• Point of Contact: Nicole Jackson, Acting Director, OCPO Procurement Policy and Systems, is the point of contact for this instruction.

Form / Attachment / Pricing Sheet

This document outlines the "AS-2602 Privacy, IP, and Government Data Use Requirements" clause. It details the mandatory compliance requirements for contractors handling Federal information, particularly concerning privacy, intellectual property (IP), and the use of government data, including Artificial Intelligence (AI) Systems.

• Compliance: Adherence to the Privacy Act of 1974, Federal Records Act, E-Government Act 2002, and relevant HUD policies.
• Intellectual Property & Data Ownership: Government retains ownership of data, queries, prompts, and outputs generated by AI Systems. Contractor use of non-public government data for training AI is prohibited without explicit written consent.
• Access Rights: Government must have access to all AI system components for monitoring and evaluation.
• Data Handling: Strict protocols for collecting, accessing, retaining, and protecting Federal records, including compliance with NARA regulations and safeguarding sensitive information.
• Record Management: Contractors must prevent unauthorized destruction or alienation of Federal records and report any incidents.
• Subcontractor Compliance: Requirements must be flowed down to subcontractors involved with AI systems or data.
• Data Transfer: Data transfers must use a Secured File Transfer Protocol.

This document is critical for bidders to understand the stringent requirements regarding data privacy, intellectual property, and the handling of government data, especially when AI systems are involved. It outlines the contractor's responsibilities and potential liabilities related to data security and compliance, which will impact proposal development, contract execution, and overall cost.

Statement of Work (SOW) / Performance Work Statement (PWS)

This document outlines the requirements for ongoing testing and monitoring of Artificial Intelligence (AI) Systems or Services throughout their lifecycle to ensure continued performance, accuracy, effectiveness, and trustworthiness.

• Contractors must support regular monitoring and evaluation of AI Systems.

• Evaluations may include assessments of performance, accuracy, fairness, bias, security, explainability, and overall fitness for purpose.

• The Government reserves the right to monitor, test, and evaluate AI Systems at various intervals (quarterly, biannual, or other).

• The Government may use internally defined datasets for independent evaluation, which will be selected and maintained solely by the Government.

• Contractors must provide necessary access and technical support for Government AI System evaluations.

• Contractors must provide test results, methods, and documentation to allow independent reproduction or verification of findings.

• Contractors must permit Government oversight of the AI System evaluation process.

• Contractors shall disclose AI System examination and validation procedures, documentation, and results.

• Not explicitly detailed, but evaluations will assess performance, accuracy, fairness, bias, security, explainability, and fitness for purpose.

• Ongoing throughout the AI System's lifecycle.

• Contractors must flow down these requirements to subcontractors providing or utilizing AI components/systems.

• Contractors must provide an estimate of the cost to bring AI Systems/Services into compliance.

• Failure to adhere to these requirements may result in contract remedies, including stopping work until compliance is demonstrated.

Form / Attachment / Pricing Sheet

This document, "AS-2603 AI Use Transparency Requirements," outlines specific requirements for contractors regarding the disclosure and use of Artificial Intelligence (AI) systems in products and services delivered under a contract. It serves as a mandatory clause for solicitations and contracts above the micro-purchase threshold.

• High-Impact AI Disclosure: Government may identify AI systems as "High Impact AI" during solicitation or contract performance.
• Vendor Transparency Obligations: Contractors must obtain written authorization from the Government before using any AI system. This includes submitting detailed documentation for each AI system, such as design, intended function, operational context, compliance with Unbiased AI Principles, system context diagrams, data flow diagrams, and more.
• Unbiased AI Principles: The document references Executive Order 14319 and mandates compliance with "Truth-seeking" and "Ideological Neutrality" principles for AI systems, particularly Large Language Models (LLMs).
• Documentation Requirements: Extensive documentation is required, including Software Bills of Materials (SBOMs), details on human oversight for automated decision-making, risk mitigation strategies (bias, privacy, explainability), AI Assurance Packets, and AI Impact Assessments.
• Monitoring and Reporting: Contractors must continuously monitor AI system use and performance, report new AI system features, and ensure subcontractor compliance.
• Prohibitions: Contractors are prohibited from using publicly hosted AI systems for HUD data, retaining HUD data for training by third parties, introducing unapproved AI systems, or deploying autonomous AI systems without human validation.

Bidders must understand these requirements to accurately estimate costs, develop compliant proposals, and plan for contract execution. Failure to comply with these AI transparency requirements is a material requirement for contract award and can lead to significant consequences, including contract termination or damages.