Better ways to fill vacancies, AI and voting rights, and timely links

Welcome to the latest edition of The Expand Democracy 3. We typically release every two weeks and highlight promising pro-democracy ideas and local, national, and global news. And don’t miss our complimentary Democracy Lab podcast with a string of strong guests.

Expand Democracy is a charitable organization that depends on donations from people like you. If you’re enjoying this newsletter, please share it with a few people who might find it interesting. Thank you!

---

#1. Deep Dive: Better Election Methods Can Improve the Way We Fill Vacancies

By Rob Richie and Eveline Dowling

New South Carolina Senator Darline Graham Nordone. Source: New York Times

Sen. Lindsey Graham died suddenly on July 11, soon after returning from a trip to Ukraine. His death left South Carolina without one of its two US senators in the middle of a congressional session. Gov. Henry McMaster moved quickly by appointing Graham’s sister, Darline Graham Nordone, to hold the seat until the end of his term in January. Since Graham had won his party’s June 9 primary before he died, South Carolina law mandates Republicans hold a special primary to replace him on August 11, with a runoff on August 25.

The process to fill a congressional vacancy can be slow, inefficient, and unrepresentative. On the one hand, Graham’s seat is being filled the way most Senate vacancies are: a temporary gubernatorial appointment followed, eventually, by an election. The Seventeenth Amendment requiring elections for full Senate terms leaves the particulars for filling vacancies to the states. Most, including South Carolina, allow the governor to name a placeholder while a permanent successor is chosen at the ballot box. While gubernatorial appointments have a history of controversy (here’s looking at you, Trump-pardoned former Illinois governor Rod Blagojevich), they at least allow states to keep voting representation in the Senate.

However, the US House works differently and more strictly. Article I, Section 2 of the Constitution requires that House vacancies be filled “by writ of election.” Thus, no governor has ever had the power to appoint someone to the House and every one of the 435 seats has always been filled by a vote of the people. That constitutional guardrail is worth defending, but it becomes costly when an election, sometimes two, has to happen before a district has any representation at all. The 11 US House vacancies during this Congress have come at a time when the House majority is razor-thin and vacancies can shift the balance of power.

Vacancies in this Congress highlight those costs. California’s 1st District sat empty after Rep. Doug LaMalfa died on January 6. The seat was filled on June 2, when Republican James Gallagher won outright with a majority, avoiding the need for a runoff. In contrast, California’s 14th District had a vacancy due to Eric Swalwell’s resignation, with that election awkwardly scheduled on June 16, just two weeks after the regularly scheduled June 2 primary. Aisha Wahab – a Democrat who is the nation’s first Afghan-American woman elected to public office – led 43% to 17%, but fell short of a majority., That result kept the seat vacant and will require 14th District voters to cast yet another extra vote in an August 18 runoff - their third election in less than three months.

There’s another deeply concerning problem: Governors can use their power over scheduling the timing of the House vacancy election to accelerate or delay replacements. In Texas, for example, Sylvester Turner died in March 2025, yet the first round of a two-round election process to replace him was not scheduled by Republican governor Greg Abbott until November, with a runoff in January - 332 days after Turner’s death. This timeline was similar to the nearly nine months orchestrated by Florida’s Republican governor Ron DeSantis in 2021 to fill a House seat also made vacant by the death of a member of the Congressional black Caucus – in stark contrast to the aggressive 2025 schedule DeSantis established to fill a vacancy in a heavily Republican district in less than three months.

Based on state law, vacancies may require separate party primaries before a general election, with South Carolina having a majority requirement for the primary that may require a primary runoff. Others allow parties to nominate without a primary. Several states have special elections that are like California’s all-candidate model with a runoff if no candidate earns a first-round majority running against every other candidate,

None of these rules are mandated by the US Constitution or federal law. Having two or three rounds of voting to fill vacancies is a policy choice that costs a state’s taxpayers money and a full voice in Congress. Every runoff means another round of ballots printed, poll workers hired, and voters who have to show up twice, often with turnout falling sharply between rounds – meaning the “majority winner” the runoff produces may reflect a much smaller and less representative slice of the electorate than that of the first round. It can delay the selection of a representative to Congress for months.

There’s a proven alternative that has already been tested at the local level – a single election with ranked choice voting (RCV). Among the dozens of cities using RCV, many hold vacancy elections, including New York City, San Francisco, and Oakland. In every case, they hold a “just once” election with RCV. Hawaii has adopted this model for any future congressional vacancies, and California reformers are generating interest in the idea.

Even South Carolina will benefit from RCV next month. Along with five other southern states, South Carolina election law establishes that overseas voters cast RCV ballots when returning their ballots. Those RCV ballots are then counted in the runoff for whichever candidate is ranked higher on the ballot. As a result, South Carolina can hold a runoff just two weeks after the first round instead of the 10 weeks that Texas takes for its runoffs to accommodate overseas voters.

Maine is also providing lessons in the value of RCV, Maine Democrats just held a highly successful primary with RCV that nominated unifying candidates, inspired record voter turnout, and avoided time-consuming, divisive runoffs. But after their Senate nominee Graham Platner withdrew from his campaign against Republican incumbent Susan Collins, the state party had only two weeks to nominate a replacement from a field of seven candidates at a July 25 convention.

As Steven Hill lays out in DemocracySOS, Maine Democrats chose a “sequential runoff,” with repeated rounds of voting involving dropping the last-place finisher and then having convention delegates vote again rather than hearing from the candidates and casting a single RCV ballot. Washington Post columnist Henry Olsen aptly observed, “The selection process for the Senate nominee almost guarantees hard feelings will erupt. Rather than hold one round of ranked-choice voting, the party has decided to use multiple ballots… That system is ripe for dealmaking between rival campaigns, especially if the field is winnowed to three. It’s not implausible that two candidates could try to team up if they believed the third to be too weak or too moderate. Such backroom deals were staples of the old convention system.”

Moving forward, Congress should act on its constitutional power to regulate vacancy elections -prohibiting gubernatorial “date-mandering” and creating incentives to fill seats more quickly with RCV. It also should anticipate more clearly a national tragedy where scores of House seats become vacant and threaten continuity in Congress. States should also enact such laws.

In sum, unexpected deaths and sudden vacancies in elected office have consequences that we are watching continually unfold.The underlying design question about how fast and how decisively a democracy can replace a lawmaker without depriving voters of a say in the outcome isn’t new. Neither is the fix.

---

#2. Spotlight: Democracy in the Age of AI

By Nivea Krishnan

Source: The Economist

The question of how to regulate artificial intelligence (AI) is ubiquitous across academic and social circles today. From higher education to corporations, there seems to be a concerted effort towards instituting best practices. And yet, little discourse addresses the less-visible, corrosive way that ungoverned AI usage hinders the pursuit of a racially representative democracy.

This is the problem George Washington law professor Spencer Overton confronts head-on in his paper, Overcoming Racial Harms to Democracy from Artificial Intelligence updated in January 2025. His thesis is that, left unchecked, generative AI will exacerbate existing racial inequalities and empower bad faith actors to perpetuate racially driven attacks that hinder the democratic participation and civic engagement of people of color.

To be clear, Overton does not believe these consequences are inevitable; rather, they must be proactively combatted with an AI governance structure that centers democracy, not just productivity. He wholeheartedly recognizes, as do most organizers, just how powerful a tool AI can be in bringing communities together, increasing the public’s understanding of policy matters, and empowering civic participation. Realizing these positive impacts, however, requires identifying and mitigating the technology’s negative effects.

Overton’s paper underscores the myriad of ways in which AI can shape public opinion and dictate the participation of communities of color. While AI does not create unique problems, it deepens existing ones such as racial polarization, cultural anxiety, anti-democratic attitudes, unequal political representation, and barriers to voting. Overton connects these negative effects to both AI’s usage and design.

The weaponization of disinformation predates the ascension of generative AI, but the facility with which an inexperienced user can generate falsities with ChatGPT or Claude only heightens disinformation’s detrimental effect.

To quote Overton’s example of a coordinated disinformation campaign from the 2016 election:

For example, on Election Day 2016, the operators of the Williams & Kalvin Facebook page—ostensibly two Black men from Atlanta who ran a popular Facebook page focused on Black media and culture—paid for and posted a Facebook ad targeted at Black users. The ad proclaimed: “We don’t have any other choice this time but boycott the election. This time we choose between two racists. No one represents Black people. Don’t go to vote.”91 After the November 2016 election, an investigation revealed that the Williams & Kalvin Facebook account was fake and had been set up and operated by the Russian Internet Research Agency (the “Russian Agency”).

Generative AI models are now considered commonplace, readily available to any user regardless of their level of expertise or financial resources. While Overton recognizes the benefits of improved accessibility, he notes how this same access can empower bad actors to launch more coordinated, less detectable, and significantly more compelling disinformation campaigns wherever and whenever they choose. Users can now generate a broad range of fake content, from synthetic audio clips to videos to hundreds of social media accounts claiming to represent distinct, authentic perspectives, all at a fraction of the cost. Such acts, he states, will continue to be targeted at communities of color which will disrupt the deliberative process that constitutes a key element of any functioning democracy.

This proliferation of disinformation campaigns has tangible effects on the democratic participation of racial minority groups. For one, such campaigns deepen existing feelings of distrust that communities of color have for American democratic institutions, decreasing the inclination for such communities to engage in the political process. For another, the widespread acknowledgment of the presence of disinformation online chills demands for social justice.

Overton writes:

Widespread use and public awareness of deceptive synthetic content willcreate new challenges, such as skepticism, disengagement, and callousness about racial justice issues. For example, Black teenagers who witness and record on video the unwarranted killing of their friend by a police officer may face allegations that the video is a deepfake. They may face greater hurdles in generating support—particularly across racial lines—to hold the officer accountable and enact public policy change.

The design of AI models also propagates racial biases and stereotypes. These models are programmed to generate outputs based on the synthesis of inputted data, an iterative process informed by developer-curated training datasets. Overton specifies the inherent problem here:

Because foundation models are optimized to reflect their training data as accurately as possible to detect patterns, we should expect them to “reinforce stereotypes and unfair discrimination by default.”162 Training datasets may be biased because they are gathered from contexts that reflect societal inequality and/or because some communities are better represented in training data (an AI image generator trained on portraits and photos of the forty-four white males and one Black male who have served as U.S. presidents, for example, would likely not produce an image of a Black female U.S. president)

AI has trouble recognizing those groups who are underrepresented in training datasets, making it a questionable tool to apply in democracy contexts. The paper highlights signature verification as an example:

An AI program to verify signature matches that is trained on a dataset of signatures from a relatively homogenous state may have higher error rates when used to verify the authenticity of signatures on mail-in ballots in a relatively diverse state with large shares of Asian and Spanish surnames. Voice-assistance technology used to gather and process political information and to actually cast a ballot at the polls may be more accurate in standard varieties of English than in other languages or nonstandard dialects of English.

By outlining the corrosive potential of unchecked generative AI usage on a democracy, Overton makes a compelling case for what timely, comprehensive AI regulation must look like. First, policymakers must explicitly focus on the impact of AI systems on democratic participation the same way they evaluate its impacts on innovation, economic competitiveness, and privacy. Next, governance must account for the disproportionate harms AI can pose to communities of color, evaluating a system’s capacity to, for example, suppress voter participation or spread targeted disinformation. Overton also emphasizes the need to expand civil rights protections to address AI-related harms, recognizing how algorithms can create new forms of discrimination in democratic processes.

Finally, Overton clarifies that the project of AI governance is a societal one and must be a priority for policymakers, researchers, business leaders, election officials, and technology companies alike. His paper draws attention to the fact that defending democracy requires actors to recognize it as its own policy issue, in need of proactive and coordinated action.

Read Spencer Overton’s full paper here.

---

#3. Timely Links and Shorts

• “Why did the President of the United States give a prime time speech on election integrity?” Mike Latner writes in Democracy Cookbook, “[In 2020, Donald Trump] used the law until he couldn’t. At each step — in the courts, in the state capitals, and on the evening of January 3, 2021, when Jeffrey Rosen, Richard Donoghue, and Steven Engel told him from the Oval Office they would resign en masse rather than sign — the institutions of government brought him to heel. He did not concede the argument. The argument was overruled by the machinery around him. The difference this time is not the argument. The difference is that the Justice Department, ICE and the Department of Homeland Security, and the US surveillance apparatus is behind him now rather than bringing him to heel.”
  

• “FAQ: Voting system vulnerabilities and how they can be weaponized”: Geoff Hale writes for the Center for Democracy and Technology, “Security researchers do find real weaknesses embedded in the code and operations of U.S. voting systems, just like they do with power grids, banks, and telecommunications, all of which share the same “critical infrastructure” designation as elections. Finding a flaw and fixing it is a sign of a mature system, not a sign it’s broken. But claiming that vulnerabilities in voting systems exist is different from claiming that vulnerabilities in voting systems have been exploited, and even further distinct from claiming that vulnerabilities in voting systems have been exploited during live election operations in a manner to have rigged past elections, flipped votes, and determined outcomes.”
  

• Trump’s firing of Election Commissioners illustrates America’s crisis of election governance: Kevin Johnson and Heather Balas write for the Guardrail, “Yesterday, President Trump fired all the remaining members of the Election Assistance Commission (EAC), the federal agency responsible for guidance and support to state election administration. This dangerous attack on an independent entity established by Congress was enabled by the Supreme Court’s misguided decision to overturn precedents protecting such agencies from presidential control.Some will speculate about other agencies President Trump might target and debate this expanded presidential power. But, to us, his action illustrates another critical problem that isn’t talked about enough – America faces a crisis of election governance.”}
  

• NAACP LDF president - “We need a re-founding”: Janai Nelson writes in The Contrarian, “We must create institutions that are accountable to the people — including a Supreme Court that is bound by exacting ethical standards and unwavering fidelity to justice and the rule of law. This starts with reforms like term limits, curbing the abuse of the emergency docket, and a close examination of the merits of court expansion.We must overhaul our election system by expanding vote-by-mail options, limiting aggressive purges of voter rolls, codifying same-day and automatic voter registration, and ending the pernicious practice of racial and political gerrymandering. And we must begin socializing the use of alternative voting systems like ranked choice, instant runoff, and cumulative voting, which, with necessary investments in turnout, give voters more options to exercise their power to elect representatives of their choice.”
  

• “How Americans are engaged with news, politics, religion and civic Llife”: A new Pew Research Center study from the Pew-Knight Initiative sought to understand how Americans’ engagement with each other looks across a range of behaviors. Results show adults sort into four distinct groups that reflect different patterns of participation: mobilizer, connector, spectator, and outsider. Each group has different behavioral engagement with political activity, civic involvement, religious attendance, and attention to news.

• “As lawmakers politicize the 250th anniversary, Americans are looking for unity over division”: A new article from the Conversation by Tatishe Nteta and Adam Eichen from UMass Amherst reports, “...63% agreed that “The U.S. is one nation with a shared history and common future, and should never be divided,” including strong majorities from both parties. In the midst of serious challenges at home and abroad – including the Iran war, a weak economy and the conflict over immigration – we found that Americans strongly believe there is more that unites Americans than divides them.”

• “Ranked choice voting, campaign spending, and independent expenditures: An exploratory comparative analysis”: A new study on RCV, independent expenditures and campaign spending by Jeanne Clelland and Kevin Clelland was published on SSRN. It examines the relationship between campaign expenditures and electoral performance in RCV elections compared to plurality elections and finds campaign spending is a weaker predictor of electoral performance under RCV than under plurality voting.
  

• “What three polling paces in Washington, D.C. taught me about ranked choice voting”: Ed Shoemaker of Voter Choice MA writes in The Renovator, “In June, I went to Washington to strengthen relationships with national partners working to improve democracy, meet with members of the Massachusetts congressional delegation (my home state) about the Fair Representation Act, and spend time with election officials, organizers, researchers, advocates, and volunteers from across the country.I expected those meetings to define the trip. They didn’t. When people ask about the trip, they usually ask about Capitol Hill or the meetings I had. But the conversations I can’t stop thinking about took place outside three neighborhood polling places.”
  

• “Largest housing affordability bill in decades becomes law without Trump’s signature”: In a now-rare bipartisan policy win that reflects how Congress often used to operate, NPR reports, “The housing law aims to make homeownership more affordable, primarily by encouraging homebuilding across the country. It’s packed with more than 40 provisions, contributed by both Republicans and Democrats, on everything from corporate home ownership to manufactured home construction, which helps explain why it passed with overwhelming bipartisan support.”