When President Biden withdrew from the presidential race, many people asked whether it was too late for states to put a new Democratic nominee on the ballot. This wasn’t a hard question: as countless election officials and experts have explained, Biden wasn’t actually the nominee yet, so there would be no legal or practical impediment to his replacement, Vice President Harris, appearing on the ballot once formally nominated.
But Grok, an artificial intelligence chatbot available on X (formerly Twitter), was falsely telling users that the deadline had already passed in several states, as reported Sunday in the Washington Post.
A group of state election officials in the impacted states called on X to implement changes to prevent Grok from providing inaccurate information to voters. Grok is far from the only purveyor of misinformation; other AI chatbots have generated false or misleading information when users asked how or when to vote or have refused to say who won the 2020 election. But as the election officials pointed out, other tech companies have worked to correct or mitigate inaccuracies, including by directing users who ask certain election questions to nonpartisan, authoritative resources like CanIVote.org.
While these measures are far from sufficient, they represent an important contrast to X’s approach — the company’s chatbot repeated the false claims for more than a week before the issue was finally corrected.
Sadly, this generative AI debacle is not the sole instance of misleading election information that X owner Elon Musk is associated with this year.
Last Friday, CNBC reported that a Musk-funded super PAC — America PAC — spent more than $800,000 on ads, many of which were apparently designed to collect voter data, sometimes on false premises. The ads encouraged people to visit America PAC’s website and implied that they could use this site to register to vote. According to published reports, while users who entered zip codes from noncompetitive states were directed to their state’s registration website, users indicating that they lived in Georgia, Michigan, Pennsylvania, or another swing state were instead directed to a form on America PAC’s site to enter more personal information, including their address, mobile phone number, and age. After submitting the form, a “thank you” page appeared that offered no assistance in actually registering to vote.
Federal law prohibits conspiracies to prevent someone from voting. Many states protect voters against certain kinds of deception. Indeed, state officials in Michigan and North Carolina are investigating whether America PAC violated state law. (Perhaps in response, the website no longer has a link for registering.) It’s not clear that America PAC’s website violates any federal or state law. However, it made it harder for election officials to promote accurate election information and undermined access to voting. If someone who visited the website mistakenly thinks they have registered to vote by submitting their information, they may not be able to vote when they arrive at their polling place and learn that the deadline to register has already passed.
The spread of election falsehoods — including, most prominently, denial of the 2020 presidential election results — continues to threaten American democracy, spurring restrictive voting laws, attacks on election workers, and declining confidence in elections.
Ultimately, the best way to protect eligible voters from deceptive practices and other efforts to keep them from voting is for Congress to pass the Freedom to Vote Act and the John R. Lewis Voting Rights Advancement Act, comprehensive bills that set baseline national standards for voter access — including automatic and same-day voter registration and stronger, uniform protections against voter deception and intimidation. The Freedom to Vote Act also includes a slate of campaign finance reforms to rein in groups like America PAC that can raise unlimited money from billionaire megadonors and often work closely with candidates. These bills came close to passing in the last Congress and must be a top priority in 2025.
There are also targeted measures Congress and state legislatures can adopt to deal specifically with the problem of deceptive AI in elections. Legislators should require generative AI content to carry watermarks and mandate that campaigns and PACs disclose when they deploy generative AI technology to engage voters. A new Brennan Center analysis shows that state legislatures have been passing bills to regulate AI, but these efforts have focused almost exclusively on the potential for deepfakes to be used in political communications. Future bills must prioritize voters’ interests and tackle the broader harms caused by false information that AI technology enables.
And as this past week demonstrates, tech companies that develop and profit from AI must take stronger action to prevent their tools from misleading voters. Among other things, they should ensure that chatbots direct users to authoritative sources of election information, embed AI-generated content with markers that allow the public to track its origins, and create and enforce rules against use of their tools to deceive voters or suppress votes.
Finally, in the face of government inaction and the irresponsibility of some tech platforms, it is even more imperative that voters use authoritative sources to understand election processes, learn more about voting options in their state, and confirm that they are eligible and registered to vote. The Brennan Center and others are working hard to provide accurate election information and protect American democracy.
