Learning Objectives

By the end of this lecture, you will be able to: (1) distinguish internal AI governance bodies (not subject to the Federal Advisory Committee Act) from external advisory committees (subject to FACA at 5 U.S.C. Appendix); (2) design a FACA-compliant AI advisory committee with balanced membership, chartered status, public meetings, published minutes, and conflict disclosures; (3) understand the National AI Advisory Committee (NAIAC) established under the National AI Initiative Act of 2020 and chartered under FACA; (4) operate internal AI governance structures including Chief AI Officer, CAIO Council (established by OMB M-24-10), Chief Data Officer Council, Senior Agency Officials for Privacy, General Counsels, and agency Civil Rights Officers in coordinated review; (5) coordinate with external partners including the Coalition for Health AI (CHAI), the Federal Committee on Statistical Methodology, MITRE ATLAS, the NIST AI Safety Institute Consortium, and agency-specific bodies; (6) operate independent oversight structures including agency Inspectors General, GAO, and PCLOB where applicable; (7) apply FACA requirements for chartering, membership balance, closed meeting exceptions, subcommittee governance, and the General Services Administration Committee Management Secretariat role; (8) integrate ethics-board output with OMB M-24-10 minimum practices, NIST AI RMF functions, and agency AI Strategy; and (9) interpret real-world examples including the VA AI Oversight Committee, the DOD AI Principles implementation, the Axon AI Ethics Board experience, and lessons from the Google AI ethics board dissolution.

Key Topics Covered

Topic 1: FACA basics. The Federal Advisory Committee Act (5 U.S.C. Appendix) governs federal advisory committees. Requirements: chartering (with the GSA Committee Management Secretariat), balanced membership representing points of view, public meetings with reasonable notice in the Federal Register, publicly available minutes, conflicts disclosure, and sunset provisions. Subcommittees report through the parent committee.

Topic 2: FACA exemptions. Internal federal meetings (all members are federal employees on official duty). 'One-shot' interactions that do not constitute an advisory 'committee.' Certain statutory exemptions. The GSA Committee Management Secretariat and agency Committee Management Officer issue guidance.

Topic 3: Internal AI governance structures. OMB M-24-10 established the CAIO Council. Agency-level structures include: the CAIO (required by M-24-10); internal AI Governance Council or Board; Chief Data Officer (under the Foundations for Evidence-Based Policymaking Act of 2018); Senior Agency Official for Privacy; Chief Information Security Officer; General Counsel; Civil Rights Office; OGC; Human Capital Office; Acquisition Office.

Topic 4: External advisory committees (FACA). The National AI Advisory Committee (NAIAC), chartered under the National AI Initiative Act of 2020, advises the President on AI. Agency-level examples: VA AI Oversight Committee (with appropriate FACA structure); NIH Advisory Committee to the Director; DOD Defense Innovation Board AI subcommittee.

Topic 5: External partners (not advisory committees). Coalition for Health AI (CHAI), multi-sector coalition on health AI assurance. Federal Committee on Statistical Methodology. MITRE ATLAS for adversarial ML. NIST AI Safety Institute Consortium (launched February 8, 2024, with 200+ members).

Topic 6: Oversight structures. Agency Inspectors General. GAO. Privacy and Civil Liberties Oversight Board. Foreign Intelligence Surveillance Court (for FISA-related matters). DOJ Civil Rights Division and agency civil rights offices. Congressional oversight committees.

Topic 7: Design principles for effective boards. Clear charter and scope. Balanced membership including affected communities where appropriate. Substantive staff support. Documented deliberation and recommendations. Responsive agency action. Published outcomes. Periodic review and renewal.

Topic 8: Real-world examples. VA AI Oversight Committee, clinical AI oversight with public listening sessions. DOD AI Principles implementation via CDAO and Defense Innovation Board input. Axon AI Ethics Board, 2018-2021, dissolved after public pressure on facial recognition. Google AI ethics board (ATEAC) dissolved in 2019 within days of launch. Lessons: legitimacy requires substantive member diversity and agency follow-through.

Topic 9: Integration with M-24-10. The CAIO operates internal governance. Waivers under 5(d) require public notice via inventory. Agency AI Strategy under M-24-10 includes governance description. Advisory committees (FACA) advise the agency on strategy; ethics boards review specific systems; oversight bodies audit.

Why This Matters for Government

Ethics boards and advisory structures matter because AI is too consequential to be left to single decision-makers and too technical to be governed by general executive authority alone. A federal agency that deploys rights-impacting or safety-impacting AI without appropriate governance structures is one bad decision away from a GAO finding, an OIG audit, a Congressional hearing, or a civil rights investigation. The structures are the mechanism by which diverse expertise and perspectives enter the decision process in time to prevent harm rather than explain it afterward.

The historical record is informative. Consider Google's ill-fated ATEAC (Advanced Technology External Advisory Council), announced March 26, 2019 and dissolved April 4, 2019, nine days later, after public controversy over the membership composition. The lesson is not that ethics boards are doomed; it is that legitimacy requires substantive diversity, chartered authority, and genuine agency willingness to follow advice. Axon's AI Ethics Board, established 2018, resigned en masse in June 2022 after Axon announced plans to equip Tasers with drones following the Uvalde school shooting; the board had advised against the direction. The lesson: ethics boards that are ignored lose their function and their public credibility. Federal agencies face versions of both failure modes.

Positive examples exist and they share patterns. The VA AI Oversight Committee conducts real review with published outputs, listening sessions, and feedback incorporation into clinical AI programs. The National AI Advisory Committee, chartered under the National AI Initiative Act of 2020 and renewed under EO 14110, has issued substantive reports on workforce, safety, and strategic AI issues. The CAIO Council, established by OMB M-24-10, coordinates across CFO Act agencies and has produced model designation memos, minimum-practices templates, and shared tools. The Coalition for Health AI, a multi-sector body not formally federal but closely coordinated with HHS, FDA, and federal health agencies, has produced assurance laboratory pilots and published model card guidance. Each of these works because it has a clear charter, substantive diversity of expertise, dedicated staff support, and an institutional relationship that ensures its output is consumed and acted upon.

The Federal Advisory Committee Act (FACA), at 5 U.S.C. Appendix, is the statute that governs federal advisory committees. Enacted in 1972 in response to concerns about advisory bodies as 'shadow governments,' FACA requires chartering, balanced membership representing diverse points of view, public meetings with Federal Register notice, publicly available minutes, conflict-of-interest disclosures, and sunset provisions. The General Services Administration's Committee Management Secretariat oversees government-wide FACA compliance; each agency has a Committee Management Officer. FACA does not apply to meetings composed entirely of federal employees conducting official business, that is a key structural distinction between internal governance (not FACA) and external advisory committees (FACA).

The practical implication is that agencies need to think structurally. An internal AI Governance Council composed of the CAIO, CDO, SAOP, General Counsel, CISO, and Civil Rights Officer is not FACA and does not need chartering, but it also does not bring external expertise. An external advisory committee chartered under FACA brings external expertise and political legitimacy but is slower, more public, and requires sustaining resources. Most federal agencies need both. The VA model, internal clinical AI governance plus public oversight committee sessions, is a functional pattern. The DOD model, CDAO internal governance plus Defense Innovation Board advisory input, is another. The HHS model, internal leadership plus CHAI coordination and CAIO Council engagement, is another. There is no one right answer; there are patterns that work and patterns that don't.

What distinguishes effective from ineffective boards. First, a clear scope that staff can explain in two sentences. Second, members with relevant expertise and diversity of perspective, including those most affected by the AI at issue. Third, substantive staff support, a board without dedicated staff produces little. Fourth, a documented deliberation process that produces specific recommendations rather than abstract principles. Fifth, agency commitment to respond substantively to recommendations, including when the agency disagrees. Sixth, published outcomes so the public can see what was advised, what was decided, and why. Seventh, periodic external review of the board's own performance. Agencies that design for these features get durable legitimacy; agencies that skip them get ATEAC-style failures.

FACA Compliance in Detail

Chartering. Each FACA committee must be chartered, filed with Congress, and registered with the GSA Committee Management Secretariat. The charter specifies purpose, scope, duties, membership composition, meeting frequency, and termination date (usually two years, renewable). The agency Committee Management Officer prepares and files the charter.

Membership balance. The membership must be fairly balanced in terms of points of view represented and functions to be performed (5 U.S.C. App. 2 section 5(b)(2)). 'Balance' is evaluated against the committee's purpose: an AI safety board need not be 50/50 pro/anti-AI, but should include voices from affected communities, technical expertise, civil liberties perspective, industry as appropriate, and other relevant perspectives.

Member categories. Special Government Employees (SGEs) subject to federal ethics rules; Representatives who represent industry or other interests and are not subject to federal ethics rules; federal employees (who do not count toward FACA's 'committee' analysis when meeting alone). Member appointment processes vary by agency.

Meetings. Must be held open to the public with timely Federal Register notice (typically 15 days), unless a closed-meeting exception applies (e.g., Government in the Sunshine Act exceptions incorporated by reference, classified information). Closed meetings require specific written justification.

Minutes and documents. Minutes must be publicly available. Committee-generated documents are generally FACA records accessible to the public, subject to FOIA exemptions.

Subcommittees. May conduct closed internal deliberations that inform the parent committee's public deliberations. Must not effectively replace the parent committee's public deliberation.

Conflicts. SGE members file OGE financial disclosure. Representative members are not subject to OGE disclosure but agencies may adopt voluntary disclosure practices. Conflict-of-interest screening is coordinated by agency Ethics Officer.

Termination and renewal. Two-year charter term (per statute). Renewal requires new charter filing. Agencies should assess whether committee purpose is fulfilled at renewal.

Common FACA failures. Meeting with external stakeholders regularly without chartering constitutes an ad hoc 'advisory committee' and violates FACA. Hiding deliberation in subcommittees without parent public meetings violates FACA. Skipping minutes or publishing only sanitized summaries violates FACA. Unbalanced membership that represents only one point of view undermines the 'balance' requirement. Agencies that violate FACA face GAO criticism, civil litigation, and invalidation of committee recommendations.

Practical mistake modes. First, calling an internal group an 'advisory committee' in documents and then holding external stakeholder meetings under that name without chartering. Second, using contractor-facilitated convenings that effectively function as advisory committees without chartering. Third, assembling rotating 'listening sessions' that in aggregate function as a standing advisory body without the chartering and balance protections. Each is discoverable by GAO or civil society and each exposes the agency.

Designing Effective Ethics and Advisory Structures

Step 1: Clarify the function. Is this ethics review of specific systems, advisory input on strategy, oversight of programs, or standards development? Different functions need different structures. Confusing them produces weakness across all.

Step 2: Internal or external. Use internal governance (not FACA) for operational decisions requiring speed and confidentiality. Use external advisory committees (FACA) for strategic advice requiring outside expertise and public legitimacy. Use external partners (CHAI, NIST AISI Consortium, academic partnerships) for technical commons work. Use oversight bodies (OIG, GAO, PCLOB, civil rights offices) for accountability.

Step 3: Design for substantive diversity. Members from technical, policy, civil liberties, affected-community, industry, and academic perspectives. A board of all technical experts is easier to manage but misses non-technical issues. A board of all policy experts misses technical risks.

Step 4: Staff support. A board without dedicated staff produces little. Designate a senior staff director with authority to drive work. Allocate support for research, meeting logistics, public engagement, and documentation.

Step 5: Charter and scope. Write a charter short enough to explain in two sentences. List specific duties, not abstract principles. Include a sunset that forces renewal analysis.

Step 6: Deliberation process. Agenda-driven meetings. Substantive pre-meeting materials. Record of deliberations. Written recommendations to the agency. Agency response within a defined time. Published outcomes.

Step 7: Feedback loop. Agency response documents each recommendation accepted, modified, or declined, with rationale. Published. Board reviews agency response at next meeting. Cycle produces learning and accountability.

Step 8: Review performance. Every two years (at charter renewal), external review of the board's own performance. Published. Renew with improvements or terminate.

Step 9: Integrate with M-24-10. Board recommendations inform AIA content, designation decisions, minimum-practices implementation, waiver decisions, and annual AI Strategy. The CAIO is the accountable implementer.

Step 10: Build multi-board coordination. Internal Governance Council + FACA Advisory Committee + CAIO Council engagement + external partner coordination + oversight cooperation. Each has a role; they must not conflict.

Common design failures. Too narrow scope, misses important issues. Too broad scope, produces no actionable output. Too small, lacks diversity. Too large, lacks decisiveness. No staff, produces nothing. Inaccessible meetings, loses public legitimacy. No publication, loses transparency. No response process, loses board engagement. No renewal, perpetuates dysfunction. No external review, misses learning.

Case Studies

Google ATEAC (Advanced Technology External Advisory Council). Announced March 26, 2019. Eight members announced including ethicists, a behavioral economist, and a national security figure. Public controversy erupted immediately over one member's views on LGBTQ rights and another's appropriateness on the board. Dissolved April 4, 2019, nine days later. Lesson: legitimacy requires substantive diversity that holds up to public scrutiny. Hasty member selection is costly. Federal agencies should not under-resource member selection.

Axon AI Ethics Board. Established 2018. Membership included prominent ethicists, law enforcement reformers, and civil liberties figures. Published reports on facial recognition and body-worn camera AI. In June 2022, after the Uvalde school shooting, Axon announced plans to equip Tasers with drones. The AI Ethics Board had advised against this direction. Nine members resigned in public statements. The board's credibility and Axon's reputation both suffered. Lesson: ethics boards that are ignored lose function and public credibility. Agency commitment to follow advice, or to engage substantively when disagreeing, is essential.

VA AI Oversight Committee. Established and operates with substantive review of VA clinical AI systems. Includes radiologists, ethicists, veteran advocates, and technical experts. Holds public listening sessions. Publishes recommendations. VA has demonstrated willingness to pause or modify systems based on committee input. The committee sustains legitimacy across administrations. Lesson: functional design with substantive response produces durable value.

National AI Advisory Committee (NAIAC). Chartered under the National AI Initiative Act of 2020 and renewed under EO 14110. Advises the President on AI. Membership balance across perspectives. Has issued substantive reports on workforce, safety, international cooperation. Serves as a model for how a FACA-chartered AI advisory committee should function at the federal level. Lesson: chartered authority, clear mission, and substantive staff support produce useful advice.

CAIO Council. Established by OMB M-24-10. Not FACA (all federal employees). Coordinates CAIOs across CFO Act agencies. Produces model designation memos, minimum-practices templates, tooling exchanges. Permits candid peer-to-peer conversation. Has developed into an effective internal coordination body. Lesson: internal structures with clear charter and active staff can produce high-value output without FACA's overhead.

Coalition for Health AI (CHAI). Multi-sector coalition including federal health agencies, provider organizations, tech companies, academic medical centers, and patient advocates. Not a federal advisory committee. It is a coordinating body. Produces assurance laboratory pilots, model card guidance, and best-practice publications. Influences federal procurement and standards. Lesson: non-FACA partnerships can scale expertise and resources beyond what any single body could muster.

DOD AI Principles implementation. 2020 DOD AI Principles. CDAO coordinates implementation with Defense Innovation Board advisory input (DIB is FACA-chartered). Combines internal governance with external advisory. The combination has sustained through administrative transitions. Lesson: hybrid internal/external design is powerful when each component has clear role.

Federal takeaway. There is no one right structure. There are patterns, scope clarity, substantive diversity, dedicated staff, documented deliberation, responsive agency engagement, published outcomes, periodic review, that distinguish effective from ineffective. Design for the patterns; avoid the failure modes.