Operationalizing High-Impact AI Determinations: Proposal Strategy Under OMB M-25-22

With the release of OMB Memorandum M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government, the federal government rescinded and replaced legacy AI procurement guidance (such as OMB M-24-18) with a streamlined yet legally rigorous operational framework. Designed to implement executive policy promoting domestic technological dominance and operational efficiency, M-25-22 fundamentally changes how civilian and defense agencies procure AI products, models, and service-embedded systems.

For capture managers and proposal teams, the centerpiece of this directive is the explicit mandate surrounding High-Impact AI Use Cases and their mandatory disclosure within federal solicitations.

The High-Impact AI Designation: What It Means in Solicitation Scoping

M-25-22 directs agency acquisition planning teams to proactively identify reasonably foreseeable use cases for any AI system or service being procured. Crucially, contracting officers are required to disclose in solicitations whether the planned implementation meets the threshold of a high-impact use case or carries a reasonable likelihood of evolving into one over the life of the contract.

A high-impact use case encompasses artificial intelligence applications that directly affect:

• Individual rights, civil liberties, or legal outcomes
• Access to critical government services, entitlements, or financial benefits
• Safety-critical infrastructure operations and automated physical system commands
• Critical resource allocations or sensitive personnel adjudications

When a procurement crosses this high-impact threshold, the contract terms expand beyond commercial software acquisition norms. Solicitations will mandate strict compliance controls governing bias testing, continual performance drift monitoring, cybersecurity telemetry, and detailed traceability of model decision logic.

Mandatory Disclosures: Contractor Obligations in Proposal Volume II

M-25-22 warns contracting officers that federal contractors will aggressively deploy AI within contract execution. Consequently, the memorandum directs agencies to require contractors to explicitly disclose in proposals:

1. Use of Generative and Analytical AI Tools: Whether contractor personnel will use artificial intelligence to deliver work product, write software code, or analyze government datasets during performance.
2. Model Training Rights and Intellectual Property: Explicit representations that the government retains unencumbered rights to its proprietary data, preventing contractors or their cloud vendors from using federal data inputs to pre-train, fine-tune, or commercialize proprietary foundation models.
3. Domestic Sourcing Proof: In alignment with domestic preference priorities, offerors must provide transparency regarding where models were designed, trained, and hosted, ensuring alignment with emerging federal supply chain and Buy American expectations.

Failing to address these disclosures in technical proposals introduces non-compliance risk under Section L and Section M criteria. Solicitations issued on or after September 30, 2025, and options exercised on or after October 1, 2025, incorporate these mandatory evaluation factors.

Proposal Strategy: Structuring the Technical and Management Volumes

To win competitive acquisitions subject to M-25-22, proposal managers must evolve standard responses from generic declarations of algorithm performance to verified operational risk governance.

1. Construct an AI Risk Mitigation Plan (AIRMP)

When a solicitation indicates potential high-impact use cases, propose a dedicated AIRMP in the technical approach. This plan should delineate:

• Continuous verification protocols for mitigating model drift
• Clear thresholds for human-in-the-loop (HITL) intervention prior to automated action
• Mechanisms for logging and auditing inference decisions using secure, tamper-evident enclaves

2. Clearly Separate IP and Pre-Trained Weights

Agencies fear vendor lock-in and vendor intellectual property poaching. Address this directly in the data rights assertion table under FAR 52.227-14 or DFARS 252.227-7013. Clearly distinguish between:

• Off-the-shelf Commercial Models: Vendor retains standard commercial rights to core model weights.
• Fine-Tuned Weights and Government Prompts: Government retains unrestricted, exclusive rights to fine-tuning artifacts, system prompts, embeddings, and downstream outputs derived from agency records.

3. Price Continuous Monitoring CLINs Realistically

High-impact AI acquisitions cannot be sustained under conventional fixed-price software licensing assumptions. Performance tracking and bias evaluations demand ongoing cloud compute, recurring security verifications, and engineering hours. Ensure pricing models include dedicated, discrete CLIN structures for model maintenance, post-deployment safety verification, and API telemetry ingestion, preventing margin erosion during post-award performance.

By operationalizing M-25-22 requirements directly inside your technical response, capture teams transform compliance mandates into discriminators that reassure federal contracting officers that high-impact systems can be fielded securely, ethically, and without vendor lock-in.