AI_TEVV_Capability_RFI_Updated_2-4-26.pdf

UNCLASSIFIED

Defense Intelligence Agency | Artificial Intelligence Test, Evaluation, Verification, and Validation (TEVV) Capability Request for

Information (RFI)

Instructions for Completing the RFI Response Template

Please answer all questions posed in this RFI to the best of your company’s abilities. A response to each question is not necessary, however all questions should inform your overall responses and approaches. Diagrams, graphs, designs, etc. may be included in response to questions.

You may also insert one (1) cover page specific to your company with any applicable graphics, text, footers, headers, etc.

This request for information is for planning purposes only. This request is not a solicitation for proposals or quotes. Please do not submit a proposal. In accordance with FAR 15.201, a response to this RFI is not an offer and cannot be accepted by the government to form a binding contract. The Government will not reimburse respondents for any cost associated with information submitted in response to this RFI. The purpose of obtaining information is to improve the understanding of Government requirements and industry capabilities.

Respondents to this RFI may be requested to provide additional information/details based on their initial submittals. The Government reserves the right to select one, some, or none of the submissions for further information. Contractors who submit information in response to this RFI do so with the understanding that U.S. Government personnel, as well as their supporting contractors, may review their materials and/or data. The Government does not intend to rank submittals or provide any reply to interested firms.

Your response must be provided NLT 1700, 19 February 2026. Please use the RFI response template to provide your response. Any questions for clarification regarding this RFI must be submitted to ST_Proposal_INBOX@dodiis.mil by 1200, 23 January 2026.

Submission of Sensitive, Proprietary, and/or Classified Information

The Government requests that respondents avoid the inclusion of sensitive, proprietary, and/or classified information in their submissions to the maximum extent possible. However, should your response require the inclusion of such information, please adhere to the following instructions.

For Unclassified Responses:

Submit unclassified responses to ST_Proposal_INBOX@dodiis.mil with subject “Artificial Intelligence TEVV RFI.”

For Controlled Unclassified Information (CUI) / For Official Use Only (FOUO):

If your response contains information marked as CUI or FOUO, do not transmit it via standard email. Please send email to ST_Proposal_INBOX@dodiis.mil and request a secure file transfer link via the DoD SAFE (Secure Access File Exchange) service. You will receive an email from DoD SAFE with a link to which you can upload your submission securely. Ensure that all CUI/FOUO documents are properly marked in accordance with DoD instructions.

For Classified (up to TS/SCI) Responses:

If your response includes classified information, you must follow the submission instructions provided on the NSA Acquisition Resource Center (ARC) website. The ARC provides the necessary guidance and procedures for the proper handling and delivery of classified materials.

Please refer to the ARC for detailed instructions to ensure your classified submission is handled in accordance with national security protocols. Do not, under any circumstances, attempt to transmit classified information through unclassified channels, including email or the DoD SAFE service.

Protection of Proprietary Information:

The Government will protect proprietary, competition-sensitive, and company-confidential information submitted in response to this RFI. Respondents should clearly mark proprietary information (PROPIN) within their submissions. Responses will be used solely for internal government employee evaluation and planning purposes and will not be shared with contractors or outside entities without the respondent’s prior written consent, except as required by law (e.g., Freedom of Information Act requests, which will be handled in accordance with applicable exemptions for proprietary business information).

Prior Government Work and Legal Review:

Respondents citing specific prior government projects, programs, or contracts, should coordinate with those government sponsors before releasing project-specific information in response to this RFI to ensure no disclosure restrictions apply. Additionally, respondents should conduct appropriate internal legal and security reviews before submitting their responses to ensure compliance with any non-disclosure agreements, contractual obligations, or security requirements associated with prior work.

The Government shall not be liable for or suffer any consequential damages for any improperly labeled sensitive, proprietary, or classified information, which may or may not lead to unauthorized disclosures, made by respondents.

Use of Information:

Information submitted in response to this RFI will be used solely for acquisition planning, capability assessment, and strategic roadmap development purposes. Responses will not be used for intelligence collection, intelligence analysis, or any intelligence mission purposes. All information will be handled in accordance with procurement and business-related information protection protocols, not intelligence oversight authorities.

Introduction

The Defense Intelligence Agency (DIA) recognizes that the development, deployment, and operation of artificial intelligence (AI) and AI-enabled capabilities (AIEC) within the Intelligence Community (IC) and Department of War (DoW) depends fundamentally upon rigorous, comprehensive, and continuous test, evaluation, verification, and validation (TEVV) activities. As AIEC become increasingly central to intelligence operations and national security decision-making, establishing trust through demonstrated reliability, cybersecurity, safety, accuracy, robustness, and adherence to ethical and legal standards is paramount. Effective TEVV of AIEC serves as the cornerstone for building confidence among operational users, mission owners, senior leadership, and policy makers that AI systems perform as intended, produce trustworthy outputs, mitigate unintended risks, and align with established legal and regulatory frameworks. DIA seeks to implement a mature AI TEVV capability to accelerate the adoption and acceptance of transformative AI technologies throughout the Defense Intelligence Enterprise

(DIE).

Background and Motivation

DIA is exploring methods to develop standardized, scalable, and measurable TEVV metrics, as well as improve upon existing TEVV frameworks to assess technical performance of AIEC.

These metrics must also effectively communicate the trustworthiness and operational readiness of AIEC to users, senior leaders, decision makers, and policy makers who must authorize their deployment into high-consequence environments.

DIA's responsibility extends beyond internal enterprise systems to include oversight authorities across multiple DIE domains under various DIE Manager (DIEM) authorities. While DIA seeks an enterprise solution for this Agency, certain mission areas may require specialized or extended TEVV approaches. DIA recognizes the critical importance of partnering with ongoing AI TEVV initiatives across DoW components, IC elements, and other federal agencies to leverage existing investments, reduce duplication of effort, promote interoperability, and align to broader community-wide standards and best practices that are rapidly evolving.

General Requirements

Therefore, DIA seeks concepts that provide both enterprise-level standardization and the flexibility to accommodate mission-specific validation requirements, while minimizing duplicative or conflicting processes. DIA believes that effective AI TEVV must address a spectrum of system types supporting analysis, collection, operations, targeting, and business management, amongst others. Each category may require tailored TEVV approaches while maintaining consistency with overarching standards and enterprise governance. Furthermore, AIEC supporting DIA may span from development/training environments through operational deployment across multiple security domains (NIPRNET, SIPRNET, JWICS, etc..) and classification levels. Solutions must account for testing and validation requirements across this continuum, including appropriate isolation, data management, and security considerations for each domain. Lastly, DIA is particularly interested in TEVV capabilities that incorporate automation to enable continuous testing and evaluation throughout the AIEC’s lifecycle, from development through deployment and operational monitoring.

Conclusion

Any proposed capability must align with applicable policies and directives including relevant Intelligence Community Directives (ICDs), Department of Defense (DoD) Manuals (DoDMs), and DoD Instructions (DoDIs), while incorporating both traditional DoW Developmental Test & Evaluation (DT&E) and Operational Test & Evaluation (OT&E) best practices and recently-published AI-specific guidance, frameworks, and testing manuals that address the distinctive challenges of evaluating non-deterministic, data-dependent systems.

Through this RFI, DIA seeks to discover industry best practices, innovative approaches, mature methodologies, and state-of-the-art tooling to bolster its AI TEVV capabilities and address the unique requirements of intelligence mission systems. DIA invites industry partners to share their expertise, lessons learned, technical solutions, and recommended strategies that will enable the Agency to establish a robust, scalable, and sustainable AI TEVV capability appropriate for the intelligence enterprise.

Question List

A Organization Experience 1 What clearance level is your facility accredited up to?

2 Provide a brief overview of your organization, including mission, expertise, and experience in AI and TEVV for AIEC.

3 Describe your organization’s experience working with the DIE.

4 Describe your organization’s experience in developing, implementing, or operating

AI TEVV capabilities for government agencies, particularly within the DoW or Intelligence Community, citing specific examples of projects, scope, duration, and outcomes. Include examples of their use in controlled environments, if applicable.

5 What partnerships has your organization established with AI industry leaders (e.g., major technology companies, AI startups), Federally Funded Research and Development Centers (FFRDCs), University Affiliated Research Centers (UARCs), academic institutions, or other research organizations in the AI TEVV domain? How have these partnerships, if applicable, enhanced your AI TEVV capabilities?

6 What certifications, accreditations, frameworks, or industry standards does your organization leverage or align to in delivering AI TEVV services? What lessons learned from DoW/IC engagements would be applicable to DIA?

7 Describe how your AI TEVV approach or methodology has evolved or improved based on operational experience with government agencies. What specific changes have you made and why? Include examples of challenges or shortfalls you encountered and how you addressed them.

B DIA’s Mission 1 Describe your understanding of DIA’s mission, operational environment, and unique

Title 10/Title 50 responsibilities. How does this understanding inform your approach to AI TEVV for intelligence applications, and how would you address challenges unique to intelligence operations verses other DoW or federal applications?

2 How would you tailor AI TEVV approaches to address the specific requirements of intelligence analysis, including adherence to tradecraft standards such as those outlined in ICD 203 or ICD 206?

3 How would your organization communicate the importance of an AI TEVV capability to support DIA’s need to accelerate adoption of AI technologies while establishing and maintaining the trust and confidence of intelligence analysts, mission owners, and senior decision makers?

4 How would you approach AI TEVV for systems that combine intelligence data from multiple disciplines (e.g., GEOINT + SIGINT + MASINT) to produce integrated understanding of objects or activities? What specific validation challenges does multi- INT fusion present, and how would you coordinate with Intelligence Community functional manager validation processes (e.g., NGA for GEOINT, NSA for SIGINT) while enabling integrated system evaluation?

5 How would you address validation of AI systems operating at scale where human vetting of all outputs is impractical? What approaches would you recommend for establishing trust in high-volume automated systems, particularly those supporting time-critical intelligence missions?

C Frameworks, Methodologies, and Best Practices 1 Describe your recommended framework or methodology for implementing a comprehensive AI TEVV capability at DIA. Include top requirements, phases, DIA-specific organizational considerations, and how this framework has evolved based on implementation experiences with other government agencies.

2 Describe best practices for establishing and operating developmental and operational testing environments for AIEC, including considerations for data management, computational resources, information security, classification levels, and isolation from operational systems. What inefficiencies or pain points have you observed at other agencies that DIA's environment should avoid?

3 How would you structure TEVV activities across the AI lifecycle, from development and pre-deployment testing through operational monitoring and periodic re-evaluation?

4 What organizational models (e.g., centralized testing center, distributed embedded teams, hybrid approach) do you recommend for an AI TEVV capability, and what are the advantages and disadvantages of each?

5 Describe how your framework would integrate with existing specialized AI quality assurance processes across DIA’s DIEM authorities while providing enterprise-wide standardization and common infrastructure.

D TEVV Metrics 1 What metrics do you recommend for assessing AI system performance, reliability, safety, fairness, robustness, security, and other relevant attributes? How would these metrics differ across AI application types (e.g., computer vision, NLP, predictive analytics)? Which metrics have proven most meaningful and actionable in previous government projects, and which proved less useful than anticipated?

2 What metrics do you recommend for assessing AI system performance, reliability, safety, fairness, robustness, security, and other relevant attributes? How would these metrics differ across AI application types (e.g., computer vision, NLP, predictive analytics)? Which metrics have proven most meaningful and actionable in previous government projects, and which proved less useful than anticipated?

3 How would you evaluate and communicate the relative confidence, trust levels, and quality of individual data sources feeding AIEC? How would you develop metrics that distinguish system readiness for different operational use cases with varying risk profiles (e.g., situational awareness vs. targeting support vs. collection management)?

4 Describe approaches for establishing thresholds, baselines, or benchmarks for AI TEVV metrics that enable go/no-go decisions for continued development or operation and/or system deployment.

E TEVV Automation and Continuous Monitoring 1 What metrics do you recommend for assessing AI system performance, reliability, safety, fairness, robustness, security, and other relevant attributes? How would these metrics differ across AI application types (e.g., computer vision, NLP, predictive analytics)? Which metrics have proven most meaningful and actionable in previous government projects, and which proved less useful than anticipated?

2 How would you implement automated monitoring of deployed AIEC to alert stakeholders when system changes (e.g., use case expansion, new data sources, methodology changes, data drift, performance degradation, adversarial attacks, or other anomalies) require renewed validation? Including specific metrics and thresholds that would trigger intervention.

3 Describe your approach to automating test case generation, execution, and results analysis for AIEC. What role do synthetic data generation, modeling & simulation, and digital twins play in your approach?

4 What DevSecOps, MLOps, GenAIOps, AgenticAIOps, or similar practices do you recommend for integrating TEVV into AI development and deployment pipelines?

F Collaboration and Interoperability 1 What existing DoW or IC AI TEVV initiatives, programs, or capabilities should DIA partner with or leverage to reduce duplication of effort and promote standardization?

How would you facilitate these partnerships?

2 How would you ensure DIA's AI TEVV capability is interoperable with or complementary to TEVV efforts across the broader DoW and IC enterprise?

3 Describe your approach to TEVV for AIEC where validation responsibilities span multiple organizations with different authorities (e.g., system developer, IC functional managers, DIEM oversight elements, operational users). How would you enable end-to-end validation while maintaining clear accountability and avoiding redundant efforts?

4 How would your solution support both DIA-wide and mission-specific governance needs? What role-based access, workflow routing, and approval mechanisms would enable multi-level governance while maintaining enterprise visibility and standards?

5 What governance structures, communities of practice, or coordination mechanisms would you recommend to facilitate collaboration on AI TEVV across organizational and classification (air-gapped) boundaries?

G Advanced TEVV capabilities 1 Describe your organization's capabilities and approach to red-teaming AIEC, including adversarial testing, robustness evaluation, and security assessment. What methodologies and tools do you employ?

2 What red-teaming approaches or techniques have proven most effective in uncovering AI system vulnerabilities in your previous government projects? What emerging red-teaming capabilities should DIA prioritize?

3 How would you assess AIEC for potential vulnerabilities to adversarial attacks, data poisoning, model inversion, or other AI-specific security threats?

4 What research activities is your organization conducting on new and emerging approaches to AI TEVV? What promising techniques or methodologies are on the horizon, and which emerging AI TEVV research areas do you believe will be most impactful for government applications in the next 2-5 years?

5 How would you approach TEVV for emerging AI paradigms such as generative AI, agentic AI, or reinforcement or federated learning?

H Implementation and Lessons Learned 1 What workforce skills, competencies, and training programs are essential for personnel conducting AI TEVV? How would you recommend DIA develop or acquire these capabilities?

2 What are the typical timelines, resource requirements (personnel, infrastructure, funding), and critical success factors for establishing an AI TEVV capability at an organization of DIA's scale? Describe any previous implementations where initial resource estimates or timelines required adjustment and what you would do differently in planning DIA's capability.

3 What unanticipated risks, challenges, or difficulties emerged in your previous AI TEVV implementations that were not identified during initial planning or that have proven more challenging than anticipated? What aspects have proven more difficult, time-consuming, or resource-intensive than expected, and how would you proactively address these realities in DIA's capability?

4 What are your top lessons learned from AI TEVV work with government agencies that should fundamentally shape how DIA approaches establishing its capability?

Include specific examples of aspects you would design or execute differently in retrospect and their impact.

5 What assumptions about AI TEVV that were common 2-3 years ago have proven incorrect or incomplete based on operational experience? Conversely, what aspects have proven easier or more straightforward than expected? How should these insights inform DIA's approach to achieve early wins or accelerate capability development?

6 Describe your organization's process for capturing lessons learned and incorporating them into improved AI TEVV methodologies, tools, or practices. How would you establish a similar continuous improvement process for DIA?

7 If you could start over on a previous government AI TEVV project with the knowledge you have today, what would be your top five changes to the approach, implementation, or capability design?

8 What is your recommended approach for implementing AI TEVV capability in an organization where sub-organizations may already have established specialized processes for mission-specific analytics? What change management, stakeholder engagement, and process integration strategies would achieve enterprise consistency while respecting existing mission-specific expertise and avoiding duplicative efforts?