AI and Democratic Governance

Learning Objectives

By the end of this lecture, you will be able to: (1) map the interaction surfaces between AI and democratic governance in federal practice, elections administration, APA notice-and-comment rulemaking, legislative text drafting assistance, constituent-correspondence triage, public participation mechanisms, and accountability through GAO/OIG, each with specific statutory context; (2) apply the Election Assistance Commission's Voluntary Voting System Guidelines 2.0, CISA election infrastructure guidance, and FBI/CISA Joint Cybersecurity Advisories on election-related AI threats; (3) distinguish AI-generated mass comments under the Administrative Procedure Act from legitimate public participation, and implement detection and response consistent with OIRA and ACUS guidance; (4) apply the OSTP Blueprint for an AI Bill of Rights principles on Notice and Explanation and Human Alternatives to federal rulemaking processes; (5) analyze the effect of deepfakes and synthetic media on federal communications, elections, and crisis response, drawing on CISA disinformation guidance and the FBI/CISA Protected Voices initiative; (6) design agency responses to AI-generated fraud in constituent interactions consistent with NIST AI RMF 1.0 MANAGE functions; (7) implement authentic public participation including town halls, listening sessions, OSTP Requests for Information, regulations.gov notice-and-comment, and FACA-chartered advisory committees that include affected communities; (8) coordinate agency practice with the Election Assistance Commission, Federal Election Commission, Federal Voting Assistance Program, and state election directors; and (9) preserve democratic legitimacy while still benefiting from AI-driven operational improvements.

Key Topics Covered

Topic 1: AI in elections administration and security. The Election Assistance Commission (EAC) Voluntary Voting System Guidelines 2.0 adopted February 10, 2021. CISA's Election Infrastructure designation (January 2017) as critical infrastructure. CISA's #Protect2024 and prior-cycle guidance. The FBI/CISA/ODNI Joint Cybersecurity Advisories on election-related threats. State election directors and the Election Infrastructure Information Sharing and Analysis Center (EI-ISAC).

Topic 2: AI-generated mass comments in rulemaking. The FCC 2017 net neutrality docket received 22 million comments, many AI-generated or fraudulent. ACUS Recommendation 2021-1 and OIRA guidance. The Administrative Procedure Act (5 U.S.C. 553) notice-and-comment framework.

Topic 3: Deepfakes and synthetic media. The FBI/CISA Protected Voices initiative. CISA disinformation guidance. The 2024 presidential cycle's deepfake incidents including the New Hampshire primary robocall using a synthetic President Biden voice (January 2024) and the FCC's subsequent ruling that AI-generated voices in robocalls violate the TCPA.

Topic 4: AI-assisted legislative and rulemaking drafting. OLC and OMB practice. Congressional Research Service and Congressional Budget Office experimentation. Limits under the Take Care Clause, Appointments Clause, and APA.

Topic 5: Constituent service and AI. Social Security Administration and IRS call centers using AI-assisted routing and summarization. VA constituent inquiries. Risk of AI-mediated loss of direct representation.

Topic 6: Authentic public participation. OSTP Requests for Information, regulations.gov notice-and-comment, challenge.gov prize competitions, FACA-chartered advisory committees, agency town halls, tribal consultations under EO 13175.

Topic 7: Accountability institutions. Government Accountability Office, agency Inspectors General, Congressional oversight committees, and the federal courts.

Topic 8: International comparators. The EU AI Act prohibits certain manipulative practices. The UK Central Digital and Data Office Algorithmic Transparency Recording Standard. The OECD AI Principles on democratic values. The 2024 Seoul Declaration on Safe, Innovative and Inclusive AI.

Why This Matters for Government

Democratic governance rests on four hard-won capacities: free and fair elections, meaningful public participation in rulemaking, faithful representation between elected officials and constituents, and independent accountability through oversight institutions. Each has been stressed by AI. An agency AI program that ignores democratic implications is risking the substrate on which the agency depends for authority and legitimacy.

Elections. The Department of Homeland Security designated election infrastructure as critical infrastructure on January 6, 2017, in part because the 2016 intelligence community assessment documented foreign interference. Since then, CISA has expanded guidance, the EAC has issued VVSG 2.0 with explicit AI considerations, and the FBI and CISA have issued joint advisories on specific AI-enabled threats including deepfake audio in campaigns, AI-generated disinformation at scale, and AI-assisted phishing of election officials. The January 2024 New Hampshire primary robocall using a synthetic President Biden voice to discourage turnout was the first widely reported U.S. federal election deepfake, and the FCC issued a unanimous declaratory ruling in February 2024 confirming that AI-generated voices in unsolicited robocalls violate the Telephone Consumer Protection Act. Federal agencies that operate election-adjacent systems, the Federal Voting Assistance Program's overseas ballot transmissions, the State Department's passport services, DHS USCIS eligibility processes, operate in this threat environment.

Rulemaking. The Administrative Procedure Act's notice-and-comment process is a core democratic participation mechanism. AI has disrupted it. The FCC's 2017 net neutrality docket received 22 million comments, a substantial fraction AI-generated or fraudulently submitted under real names without consent. The New York Attorney General's office investigated and settled with multiple broadband industry actors in 2021 over fraudulent comments. ACUS Recommendation 2021-1, Federal Agency Rulemaking Websites, addressed how agencies should handle AI-generated comment pollution while preserving legitimate grassroots participation. Agencies that handle this well preserve the APA's core purpose; agencies that handle it badly drown in noise or unfairly exclude legitimate voices.

Representation. The SSA, IRS, and VA receive constituent contacts at enormous scale. AI-assisted routing, summarization, and response drafting can make these operations more responsive, or can insert automation between constituents and government that attenuates accountability. Design choices matter. An AI that summarizes constituent letters before a staffer reads them changes who hears what. An AI that drafts agency responses without human review changes what government is telling people.

Accountability. The Government Accountability Office, agency Inspectors General, Congressional oversight committees, and federal courts catch AI failures when agency controls miss them. GAO-24-105980 on federal AI inventories, TIGTA 2023-40-034 on IRS ID.me, the DOJ Civil Rights Division investigation opened in 2024 on Allegheny County's child welfare AI, and SSA OIG audits of algorithmic case triage are examples. Agencies that cooperate strengthen the institutions; agencies that obstruct weaken them.

Agencies need to recognize that the foundational statutes of federal democratic governance, the APA, Privacy Act, FOIA, FACA, Federal Records Act, Inspector General Act of 1978, Congressional Review Act, Hatch Act, were written before AI and are being stressed by AI. The right response is not to abandon them but to implement them in a way that preserves their purpose while acknowledging the new technological environment. NIST AI RMF 1.0, OMB M-24-10, EO 14110, and agency AI Strategy documents are the principal vehicles. The goal is not to keep government from using AI, the goal is to ensure AI use strengthens rather than erodes democratic governance.

Elections and Electoral Integrity in the AI Era

The DHS designation of election infrastructure as critical infrastructure on January 6, 2017, established federal responsibility to support state and local election administrators. CISA, through EI-ISAC and the Election Infrastructure Government Coordinating Council, coordinates ongoing support. The Election Assistance Commission operates the Voluntary Voting System Guidelines program; VVSG 2.0 contains explicit provisions for AI and machine learning components in voting systems.

AI-related threats to elections include: (1) AI-generated voice, video, or text purporting to come from candidates, election officials, or news outlets; (2) AI-assisted phishing of election officials and poll workers; (3) AI-enabled social-media manipulation at scale; (4) algorithmic cuts to voter registration, voter roll maintenance, or mail-in ballot signature verification that produce disparate impact; (5) AI-driven voter intimidation via targeted messaging; and (6) AI-assisted interference with election infrastructure through novel cyber vectors.

The January 2024 New Hampshire primary robocall was the first high-visibility U.S. federal election deepfake. A synthetic President Biden voice encouraged recipients not to vote in the primary. The FCC issued a declaratory ruling on February 8, 2024, confirming that AI-generated voices in unsolicited robocalls violate the TCPA. The NH Attorney General and the FCC opened investigations. A political consultant was indicted. The prosecution demonstrated that existing federal and state laws, the TCPA, election-law voter intimidation statutes, wire fraud, and deceptive practices, already cover many AI-enabled election abuses.

Federal agency roles. DHS/CISA: support state and local election officials with cybersecurity, incident response, and threat information. EAC: voting system guidelines and election administrator training. FBI: criminal investigation of election crimes and foreign interference. DOJ Civil Rights Division: voting rights enforcement under the Voting Rights Act. FEC: campaign finance enforcement including deepfake-related misleading advertising rules. FCC: telecommunications regulation including TCPA application. FVAP: overseas and military voting. USPS: mail-ballot logistics.

Federal practice recommendations. First, agencies should avoid creating artificial high-stakes moments adjacent to election cycles. Second, agencies with constituent-facing AI should audit for election-adjacent risks in the ninety days preceding any federal election. Third, agencies should coordinate with state election directors when federal AI actions could affect voter interactions. Fourth, federal employees must comply with the Hatch Act; federal agency AI cannot be used for partisan electoral purposes. Fifth, deepfake detection and response should be part of routine incident response, coordinated with CISA and, where criminal, the FBI.

AI in Federal Rulemaking and Public Participation

The APA's notice-and-comment rulemaking process at 5 U.S.C. 553 is a core federal democratic participation mechanism. Its purpose is informed agency decisionmaking: NPRM published, public comments, agency considers, Final Rule with reasoned explanation. Supreme Court review under arbitrary-and-capricious (State Farm 1983; Prometheus Radio 2021) gives teeth to the requirement.

AI has disrupted this process in three ways. First, AI-generated mass comments at scale produce apparent public support or opposition that does not reflect actual public views. The FCC 2017 net neutrality docket of 22 million comments is the canonical case; the New York Attorney General's 2021 settlements documented millions of fraudulent submissions. Second, AI-assisted sophisticated comments, grounded, cited, on-point, create a challenge for agencies distinguishing AI from human authorship and weighting comments appropriately. Third, agencies' own AI tools for comment analysis can introduce new risks: summarization AI that miscategorizes or misquotes comments, clustering AI that suppresses minority views, or sentiment AI that misreads nuance.

ACUS Recommendation 2021-1, 'Federal Agency Rulemaking Websites,' addresses part of the problem. OMB OIRA guidance and individual agency practice, FCC, NRC, EPA, USDA, have developed approaches including: dedupe and form-letter clustering with disclosed methodology; verification challenges for new commenters; APIs for organized comment campaigns that preserve integrity; and post-comment-period analysis that acknowledges mass submissions without discounting legitimate grassroots participation.

Federal agency best practice. First, follow ACUS Recommendation 2021-1 and OMB OIRA guidance. Second, disclose your AI-assisted comment-analysis methods. Third, distinguish unique substantive comments from mass-produced form letters in the Final Rule's reasoned explanation without dismissing form-letter participants as illegitimate. Fourth, preserve raw comment data consistent with the Federal Records Act for GAO and judicial review. Fifth, coordinate with the FBI or DOJ when submissions appear to involve identity fraud. Sixth, design the NPRM publication and comment window with awareness that AI-generated submissions will occur.

Beyond notice-and-comment, agencies participate in democratic governance through: OSTP RFIs (the AI Priorities RFI of July 2023 received 10,000+ responses); listening sessions and town halls; tribal consultations under EO 13175; FACA-chartered advisory committees; challenge.gov prize competitions; and the Paperwork Reduction Act clearance process. Each has a role and a constraint, and AI is changing them all. The common principle: AI can amplify agencies' capacity to hear the public, but only if the design preserves authenticity and minority voice.

Implementation Playbook

Step 1: Audit for democratic-governance interactions. Walk through the agency AI inventory and identify every system that touches elections, rulemaking, constituent correspondence, legislative drafting, public participation, or accountability artifact production.

Step 2: Apply relevant frameworks. For election-adjacent systems: CISA guidance, EAC VVSG 2.0, FBI/CISA joint advisories. For rulemaking systems: ACUS 2021-1, OMB OIRA, agency APA practice. For constituent correspondence: Privacy Act, E-Government Act PIA, OMB M-24-10 rights-impacting analysis. For legislative drafting: agency OGC and OLC coordination.

Step 3: Integrate with M-24-10 safeguards. Democratic-governance AI systems are generally rights-impacting. Apply the full minimum practices.

Step 4: Coordinate externally. State election directors and EI-ISAC for election-adjacent work. OSTP, OMB, GAO, OIG, Congressional committees for rulemaking and accountability. Civil society (ACLU, EPIC, Brennan Center, Center for Democracy and Technology) for participation design. Peer federal agencies through CAIO and CDO Councils.

Step 5: Build detection and response. Deepfake detection tied to incident response. Mass-comment detection tied to rulemaking. Deception detection for constituent contacts. Named response protocol and escalation to CISA, FBI, and Public Affairs.

Step 6: Preserve the record. Federal Records Act. APA record preservation. FOIA-ready artifact library. GAO-auditable decision logs.

Step 7: Publish and participate. Use regulations.gov, challenge.gov, OSTP RFIs for authentic participation. Town halls, listening sessions, tribal consultation under EO 13175.

Step 8: Train operators. Staff interacting with democratic-governance-adjacent AI should receive training on Hatch Act, APA, Privacy Act, and Federal Records Act in addition to M-24-10 general training.

Step 9: Review and learn. Quarterly CAIO review. Annual AI Strategy democratic-governance section. Incident lessons fed back.

Step 10: Guard against drift. Democratic-governance erosion is gradual. Build in explicit tripwires, public trust polling, civil society engagement metrics, OIG/GAO findings, Congressional testimony outcomes, that signal when drift is occurring and require intervention.