Learning Objectives

By the end of this lecture, you will be able to: (1) prepare and deliver briefings, presentations, and congressional testimony on federal AI policy grounded in OMB M-24-10, Executive Order 14110, NIST AI RMF 1.0, and agency-specific programs; (2) tailor messages for distinct audiences including Congress, GAO and OIG auditors, OMB Desk Officers, interagency councils such as the Chief AI Officer Council and Chief Data Officer Council, agency program managers, civil society groups, journalists, and the general public; (3) comply with agency public affairs clearance processes, Office of Legislative Affairs coordination, and the Hatch Act when speaking about federal AI programs; (4) use the federal AI inventory, model cards, and AI Impact Assessments as primary evidence rather than vendor marketing materials; (5) handle difficult questions about harms, incidents, and failures, including the IRS ID.me episode, Michigan MIDAS, Dutch toeslagenaffaire, and Allegheny County investigations: with accuracy and appropriate humility; (6) meet Section 508 accessibility standards for presentations including captioned video, accessible slides, and plain-language handouts under the Plain Writing Act of 2010; (7) coordinate with the agency Senior Agency Official for Privacy, General Counsel, Civil Rights Office, and Public Affairs to avoid inadvertent disclosure of Controlled Unclassified Information, classified material, or Privacy Act-protected information; (8) distinguish personal speech (permitted with disclaimers) from official speech (subject to clearance) and avoid creating apparent agency positions in personal capacity; and (9) follow Federal Advisory Committee Act requirements when speaking as part of a chartered advisory body and press guidance when responding to inquiries.

Key Topics Covered

Topic 1: Audience analysis. Congress (House and Senate committees, particularly House Oversight, House Science, Senate HSGAC, Senate Commerce), GAO auditors, OIG, OMB Desk Officers, CAIO Council peers, agency staff, civil society including ACLU and EPIC, journalists including Brian Krebs and Nextgov reporters, and the general public. Each audience has different prior knowledge, vocabulary, and political context.

Topic 2: Message design. The minimum unit is a thesis plus evidence plus qualifier. Example: 'Our agency's use of AI for benefits adjudication (thesis) is documented in the AI use case inventory and subject to meaningful human review by trained adjudicators (evidence), though we continue to monitor for disparate impact and have published remediation steps in our latest AI Strategy (qualifier).'

Topic 3: Evidence hierarchy. Primary: statute, executive order, OMB memorandum, NIST publication, agency-authored document, OIG or GAO report, published AIA, court opinion. Secondary: peer-reviewed research, reputable journalism. Tertiary: vendor materials (use with care; never as sole basis). Avoid unverified social media, unattributed claims.

Topic 4: Federal clearance processes. Agency Public Affairs Office, Office of Legislative Affairs, General Counsel, Ethics Officer, Senior Agency Official for Privacy. Typical clearance cycle is three to ten business days for public speech; congressional testimony requires OMB clearance and often the Office of Legislative Affairs coordination.

Topic 5: Legal and ethical constraints. Hatch Act limitations on political speech. Privacy Act restrictions on personally identifiable information. FOIA and trade secrets on vendor-provided information. Classification under EO 13526. Controlled Unclassified Information under 32 CFR Part 2002. Federal Advisory Committee Act when speaking for a chartered body. Plain Writing Act of 2010 for accessible communication. Section 508 for technology accessibility.

Topic 6: Handling difficult questions. The IRS ID.me reversal, the Michigan MIDAS settlement, the Dutch toeslagenaffaire resignation of the Rutte III cabinet, the Boeing 737 MAX MCAS crashes, and the Tempe Uber fatality are the questions that come up in every government-AI talk. Prepare accurate, sourced responses that acknowledge what went wrong without overclaiming about causes, prescriptions, or analogies.

Topic 7: Congressional testimony specifics. Opening statement (typically five minutes), written statement (longer, published in record), question-and-answer. Prepare an executive summary, primary sources in footnotes, and anticipated Q&A. Coordinate with OLA on scheduling and with OMB on interagency review.

Topic 8: Media interaction. Coordinate with Public Affairs. Distinguish on-the-record, on-background, and off-the-record. Correct errors in writing. Do not speculate about ongoing investigations, personnel matters, or classified programs.

Topic 9: Accessibility and plain language. WCAG 2.1 AA for digital. Section 508 for federal EIT. Captioning for video. Alt text for images. Plain Writing Act for public communication. Multilingual support per EO 13166 for Limited English Proficiency populations.

Why This Matters for Government

An AI program that cannot be explained to Congress, to GAO, to OIG, to the press, and to the public cannot be sustained politically. Federal AI leaders will be called to speak, in testimony, in interagency briefings, in press conferences, in town halls, in classrooms, and in academic conferences, and the quality of that speaking directly shapes whether the agency's AI program survives scrutiny, secures funding, and earns public trust.

Consider the IRS ID.me case one more time, from the speaking-and-presenting angle. The initial contract with ID.me was implemented in 2021 with narrow procurement disclosure and essentially no public speaking. When the story broke in January 2022, the IRS had no prepared spokesperson, no briefing materials, no testimony package, and no quote-ready response. Senator Ron Wyden, by contrast, had a clear message, a credible expert (Brian Krebs) as source, and immediate press availability. The political imbalance was decisive within days. Treasury Inspector General for Tax Administration report 2023-40-034 later found the IRS had not completed a Privacy Impact Assessment. Had the IRS instead opened with a public briefing before deployment, with clear messaging on the security problem being solved, the alternatives considered, the mitigations implemented, and the path to redress, the outcome would likely have been very different. The technology was the same in both scenarios. The speaking was not.

Contrast this with the ways the Social Security Administration, the Veterans Health Administration, the United States Patent and Trademark Office, and the Centers for Medicare and Medicaid Services speak about their AI programs. SSA routinely publishes Analyses of Impact and presents to Congressional committees. The VA's Under Secretary for Health has appeared before House Veterans' Affairs regarding clinical AI, with documented presentations that connect system design to veteran outcomes. USPTO leaders present model cards, evaluation results, and examiner feedback at stakeholder conferences. CMS Chief AI Officer briefings to the Medicare Payment Advisory Commission (MedPAC) convey algorithm design clearly. These programs do not always agree with their critics, and they are not immune from controversy, but they sustain political and public legitimacy because their leaders can stand up and explain what they are doing, why, what the risks are, how they are mitigated, and how the public can find out more.

Speaking well matters for three reasons beyond the immediate moment. First, the speaking creates the institutional memory that survives turnover. When a new CAIO arrives, the Congressional testimony record, the public speeches, and the interagency briefings are the documentation that shapes the new leader's priorities. Second, speaking well builds the peer network, the CAIO Council, the CDO Council, the professional societies, that provides mutual support during difficult moments. Third, speaking well produces the artifacts (testimony, speeches, Q&A records) that external researchers, advocacy groups, and Congressional staff cite; this citation economy is how federal AI policy actually evolves. Silent agencies are governed by louder ones. Agencies that speak well shape the policy landscape.

Finally, speaking about federal AI is subject to meaningful legal constraints that distinguish it from commercial speech. The Hatch Act limits political speech. The Privacy Act prohibits disclosure of personally identifiable information about individuals. FOIA and the Trade Secrets Act protect vendor commercial confidential information. Executive Order 13526 classifies national security information. The Controlled Unclassified Information framework under 32 CFR Part 2002 protects sensitive but unclassified material. Federal Advisory Committee Act applies when speaking for a chartered body. Plain Writing Act of 2010 requires plain language for communications to the public. Section 508 requires accessibility of federal EIT. An AI leader who speaks without knowing these constraints creates risk; one who speaks within them, with clearance, using the right evidence, builds the foundation for a sustainable program.

Audiences and How to Tailor for Each

Congress. Members are busy, their staffs are expert, and the legislative record matters. Opening statements are short (five minutes). Written statements are long (20-100 pages typical for complex topics). Expect questions that vary from detailed technical (from committee staff with engineering backgrounds) to political narrative (from members building a talking point). Anticipate the narrative: hearings are often scheduled around a news event or GAO report. Read the relevant GAO and OIG reports, the last three hearings on similar topics, and the likely member interests. Coordinate with the Office of Legislative Affairs on scheduling, with OMB on interagency review, with General Counsel on clearance, and with Public Affairs on messaging.

GAO and OIG auditors. These are peer professionals auditing under a defined framework (AI Accountability Framework GAO-21-519SP; OIG uses agency-specific frameworks). They want primary documents: AIAs, model cards, training records, monitoring dashboards, incident logs, override data. Provide them. Do not obfuscate. The audit is going to conclude what it concludes; your cooperation determines whether it documents a learning organization or a cover-up.

OMB Desk Officers. They implement M-24-10. They want AIA, inventory entry, minimum-practices certification, waiver documentation, annual AI Strategy. They communicate through the Chief AI Officer and CFO Council channels. Speak to them in the memorandum's own categories.

Chief AI Officer Council peers. They face similar challenges. Share templates, share lessons learned (including failures), share tools. This is the federal community of practice. Speaking candidly to peers creates the trust that enables coordinated action when a problem spans agencies.

Program managers and operators. They need actionable guidance. Speak to them in their workflow language: what changes Monday morning, what the new forms look like, what the training covers, what the escalation path is.

Civil society. ACLU, EPIC, Brookings, Center for Democracy and Technology, Data & Society, AI Now. They have expertise, they have platforms, and they have the patience to engage deeply. Brief them accurately, acknowledge concerns, answer follow-ups, and build a working relationship. The Clearview AI, Palantir, and ID.me cases show what happens when agencies don't.

Journalists. Brian Krebs, Nextgov, FedScoop, ProPublica, NYT tech, WaPo tech, Bloomberg tech. Coordinate with Public Affairs. Be clear about on-the-record versus on-background. Correct errors in writing. Respect deadlines. Build a reputation for accuracy.

General public. Plain Writing Act compliance. Multilingual support per EO 13166 where affected populations have Limited English Proficiency. Accessible formats under Section 508 and WCAG 2.1 AA. Visit libraries, community centers, town halls. The Census Bureau and IRS have traditions here. AI programs should too.

Academic audiences. Peer-reviewed venues (FAccT, NeurIPS policy track, AIES), public talks at universities. These audiences value technical accuracy and methodological rigor. Prepare citations carefully. Publish slides.

Preparing Congressional Testimony

Step 1: Understand the ask. Committee chair's letter, member interests, hearing title, witness panel. Read the last three hearings on similar topics. Read the GAO and OIG reports cited.

Step 2: Assemble the evidence base. Relevant OMB memos, EO sections, NIST publications, agency AIAs, inventory entries, court opinions, peer-reviewed research.

Step 3: Draft the written statement. Typical structure: (1) background and purpose; (2) agency approach and principles; (3) specific program(s) at issue; (4) safeguards and oversight; (5) lessons learned from incidents; (6) cooperation with oversight bodies; (7) request for specific legislative or appropriations action where appropriate; (8) conclusion. Footnote primary sources. Include appendices with technical detail.

Step 4: Draft the opening statement. Five minutes, 600-750 words. One or two headline messages, one or two examples, a forward-looking commitment. Avoid technical jargon that members' staffs will parse differently than you intended.

Step 5: Prepare Q&A. Expected questions plus answers. Include likely curveballs: 'What went wrong with Michigan MIDAS and could it happen here?' 'How do you know your system isn't another IRS ID.me?' 'What's your disparate impact rate?' Don't speculate; cite. If you don't know, commit to follow up.

Step 6: Clearance. OMB, OLA, General Counsel, Public Affairs, Ethics, SAOP. Build in time.

Step 7: Rehearse. Time the opening. Rehearse Q&A with peers who will push back. Review C-SPAN footage from recent hearings to understand the room.

Step 8: Day-of. Arrive early. Bring water. Keep responses brief and substantive. If a member is hostile, remain calm and factual. 'Senator, I appreciate the question. Let me address it in two parts...' is your friend.

Step 9: Follow-up. Questions for the record (QFRs), committee members submit written questions after the hearing; responses become part of the record. Treat QFRs as carefully as the hearing itself. Typical QFR response is two weeks.

Step 10: Archive. Hearing transcripts and QFR responses are published. Your statements shape future hearings and audits. File them well.

Known pitfalls. Overclaiming accuracy (SSA learned this in early disability AI testimony). Understating risk (several agencies have been caught doing this). Making promises you can't keep on timelines (OMB M-24-10 implementation showed this). Speaking beyond your lane (deferring to colleagues with relevant authority is strength, not weakness). Ignoring the political context. Forgetting that the written record lasts longer than the hearing.

Handling Difficult Questions

Difficult questions about government AI generally fall into four categories: the harms question ('how could this cause harm?'), the incident question ('what about case X?'), the comparison question ('isn't this just like system Y that failed?'), and the trust question ('why should we believe you?'). Prepare responses for each category.

The harms question. Answer with the AI Impact Assessment: the risks the agency identified, the mitigations in place, the monitoring, and the redress. Do not minimize. Do not overclaim. Cite GAO-21-519SP and NIST AI RMF as the frameworks.

The incident question. When asked about IRS ID.me: 'The IRS learned that transparency must precede deployment. OMB M-24-10's inventory and PIA requirements are designed to prevent a recurrence.' When asked about Michigan MIDAS: 'The failure spanned multiple safeguards: no impact assessment, no subgroup analysis, no meaningful human review. OMB M-24-10's minimum practices address each of these explicitly.' When asked about Dutch toeslagenaffaire: 'The Belastingdienst case demonstrates the cost of using a protected-class variable and failing to implement redress. Title VI and ADA Title II in U.S. law prohibit such designs; our safeguards are the operational implementation.' When asked about Boeing 737 MAX MCAS: 'That case involved override infeasibility, a HOTL pattern that collapsed into human-absent. Our safety-impacting systems are designed with effective override authority and adequate training.'

The comparison question. 'Is this like Clearview AI?' 'Clearview scraped biometric data without consent and sold to law enforcement; our system operates on data we lawfully hold under a published SORN, with documented PIA, with opt-out where appropriate, and with redress.' 'Is this like COMPAS?' 'COMPAS faced Wisconsin v. Loomis caveats on training and use; our system includes documented operator training and published subgroup analysis.' These questions are invitations to explain what makes the agency's system different and more defensible; treat them as opportunities.

The trust question. Point to the independent check: the OIG audit, the GAO report, the published model card, the NIST AI Safety Institute evaluation, the academic peer review. 'The point is not that you should trust me; the point is that an independent auditor, on the public record, can verify what I am saying.' This is the structural answer.

General rules. Don't speculate. Don't commit to timelines you can't meet. Don't minimize. Don't overclaim. Don't attack the questioner. Don't disclose what you can't. Do cite. Do defer where appropriate. Do commit to follow up in writing. Do acknowledge what you don't know.

Especially with journalists: don't lose composure on camera, don't assume anything is off the record unless explicitly agreed in advance with Public Affairs, and remember that confidentiality agreements with reporters are narrowly defined and easily breached. When in doubt, say less.

Accessibility, Plain Language, and Multilingual Support

Federal presentations must meet Section 508 accessibility standards. Slides use high-contrast color palettes, sans-serif fonts at 24-point minimum, alt text for all images, and logical reading order. Handouts include large-print and screen-reader-accessible versions. Video content is captioned with accurate timing. Live presentations include CART (Communication Access Real-time Translation) when requested under reasonable accommodation.

The Plain Writing Act of 2010 requires federal communications to the public to use plain language. For AI topics, this means: define technical terms when first used; prefer concrete examples to abstract claims; use short sentences; use active voice; use 'we' and 'you' rather than 'the agency' and 'stakeholders' where appropriate; avoid bureaucratic constructions ('in accordance with' → 'under'; 'utilize' → 'use'; 'facilitate' → 'help').

Executive Order 13166 requires meaningful access for Limited English Proficiency populations in federally conducted and federally funded programs. For AI topics, this means translation of critical public-facing materials into the top non-English languages spoken by the relevant population. HHS, SSA, IRS, USCIS, and DHS lead on this. Agencies using AI for benefits determinations must provide language access for the notice and explanation elements of M-24-10 section 5(c)(vi).

The Americans with Disabilities Act Title II and Section 504 of the Rehabilitation Act require reasonable modifications for federally conducted programs. For AI topics, this means that notices and explanations must be accessible to people with visual, auditory, cognitive, and physical disabilities, including through assistive technology.

These constraints shape how you speak. A presentation that uses only dense slides with 10-point type is inaccessible. A presentation that uses jargon without definition is inaccessible. A presentation with uncaptioned video is inaccessible. A presentation in English-only to a Limited English Proficiency audience is inaccessible. Each of these is also a legal and ethical problem beyond being poor speaking. It is an exclusion.

Ethics, Hatch Act, and Agency Clearance

The Hatch Act (5 U.S.C. 7321-7326) limits political activity by federal employees. When speaking about AI in a professional capacity, stay on policy and program; avoid partisan electoral commentary or fundraising. The Office of Special Counsel issues guidance on the line between permissible commentary and restricted political speech.

The Standards of Ethical Conduct for Employees of the Executive Branch (5 CFR Part 2635) govern outside speaking engagements. Paid outside speaking about one's official duties generally requires approval; unpaid speaking may require clearance through the Ethics Officer and Public Affairs. Personal speech must not create apparent agency positions; use disclaimers ('my personal views, not those of my agency').

Classification. EO 13526 classifies national security information. Speakers must not disclose classified information, even inadvertently. Classification markings on briefings and slides must be followed. The Information Security Oversight Office (ISOO) at the National Archives issues guidance.

Controlled Unclassified Information. 32 CFR Part 2002 governs CUI. When in doubt, consult the CUI Senior Agency Official. Law Enforcement Sensitive, Protected Critical Infrastructure Information, and For Official Use Only are common CUI categories that can arise in AI topics.

Privacy Act. Do not disclose personally identifiable information about individuals from Privacy Act systems of records except under a published routine use or with individual consent.

FOIA and trade secrets. Do not disclose vendor-provided commercial confidential information (FOIA Exemption 4) without vendor consent. In a presentation, this may mean abstracting algorithm details while still conveying the system's intended use and evaluation results.

Federal Advisory Committee Act. When speaking as a member of a chartered advisory committee, follow the charter's rules. Minutes of FACA committee meetings are public; members must disclose conflicts of interest.

Agency clearance process. Typical steps: draft → Public Affairs review → General Counsel review → Ethics Officer review → SAOP review if PII is involved → final approval. Congressional testimony adds OMB and OLA. Classified presentations add Security Officer. Budget presentations add Budget Office. The process is a few days for routine talks and several weeks for testimony. Build in time.

Penalties for non-compliance. Hatch Act violations can lead to removal. Unauthorized disclosure of classified material is criminal (Espionage Act). Privacy Act violations have civil and criminal provisions. FOIA Exemption 4 violations can lead to vendor litigation. Being wrong in public on a clearance-subject topic can end a career and embarrass the agency. Do the work upfront.