Bots and Ballot Boxes: What today’s technology means for our elections
Alejandra Montoya-Boyer
Vice President, the Center for Civil Rights and Technology
The Leadership Conference on Civil and Human Rights
The threats AI presents to elections are neither new nor unexpected. Bad actors have used social media platforms to influence U.S. elections for over a decade, using dishonest and manipulative content to spread lies, sow hate, and encourage suspicion of our neighbors. Even as advocates have spent years raising alarms, the problems persist. This is not a matter of ignorance — major tech companies have acknowledged the proliferation of disinformation on their platforms[1],[2],[3] — but of willingness to make effective changes.
Not only have platforms failed to pursue meaningful improvements, they have backtracked, abandoning key elements of content moderation. In recent years, some have shifted away from traditional moderation and fact-checking by professionals to community notes that rely on volunteers. Community notes are ineffective in addressing election falsehoods: they are too slow, have limited reach, and don’t effectively refute lies.[4] These failings are particularly dangerous when election falsehoods are part of a strategic disinformation campaign meant to sow distrust.
Not only is it imperative that the companies making and using AI technologies take responsibility for the threats AI poses to elections, it is in their best interests to do so. Injecting AI, which can reproduce and personalize content at a massive scale, into the current content moderation landscape will further weaken trust in our elections:
AI makes it harder to tell what is true. As more of the internet has become AI-generated, people have grown increasingly skeptical of online information. In the first quarter of 2026, an estimated 49.9% of articles published online were AI-generated.[5] An information ecosystem in which half of new text is produced by error-prone algorithms is not reputable. The sheer volume of AI-generated content online, on top of the chatbots’ improving ability to mimic human writing, is making it progressively difficult to know what to trust.
In this environment, platforms could develop a trustworthy reputation by increasing the visibility of accurate, helpful information. But instead, some of the most influential social media platforms have retreated, leaving the choice of which information to highlight up to opaque algorithms that reward engagement over truth. Some platforms have even deprioritized official sources by making verification status available to anyone who pays for it.
AI content moderation allows dangerous content. AI is hastening platforms’ retreat from traditional content moderation by enabling them to replace human moderators with algorithms. Even as companies tout efficiency, replacing human content moderators results in worse moderation.[6] Automated content moderators miss context and subtext that humans catch — the practice of replacing letters, words, and even phrases specifically to evade automated moderation is widespread on social media. AI moderation is even worse in languages besides English: since large language models (LLMs) are trained on content written by humans, AI is worse at moderating content in other languages.[7]
AI makes it easy to generate fake content. AI is a dangerously powerful tool for bad actors seeking to disrupt elections. Creating fake content used to be labor intensive; now, AI makes it easy. AI-generated or manipulated images, audio, and video can create a narrative out of thin air: a bad actor could easily make a fake audio clip of an election official rigging results or a deepfake video of a governor announcing last-minute voting changes. Something as simple as an AI-generated image of ICE agents at the polls could damage people’s confidence that they and their neighbors can vote safely.
Another serious concern is how easily AI can make fake versions of official websites. A cloned version of a secretary of state’s website could easily spread false voting information, and a clone of a local news site could corroborate it. A voter trying to verify a rumor could be presented with seemingly incontrovertible evidence that the lie was true.
AI makes lies travel faster and further. Beyond using AI to fake evidence, a bad actor can use AI to spread election lies faster and more efficiently. Crafting targeted political messaging for specific communities, which used to require significant time and expertise, can now be accomplished in minutes: an AI chatbot could generate an election influence strategy, write social media content, identify hyper-specific audiences, and customize content. AI agents could even autonomously purchase ads and use bots to mimic engagement. The result is a disinformation machine that is faster, cheaper, and more convincing than ever.
AI makes cyber attacks easy. One of the areas in which AI is most capable is software engineering. Even someone with no technical background could “vibe code” malware to attack election infrastructure. An attack could be as simple as phishing an election administrator with an official-seeming email, or as complex hacking voting machines. An attack wouldn’t even need to break laws: AI agents could disable official communications by overwhelming phone lines or email servers.
These threats are not insurmountable. Until there are comprehensive, civil-rights-protective federal AI safeguards, there are steps tech companies can take to ensure our elections are safe and accessible for all:
- Promote verified, legitimate sources of information, and ensure that an account’s verification status (or lack thereof) is blatantly obvious.
- Down-rank accounts with a history of spreading election lies. Terminate repeat offender accounts.
- Invest in professional human content moderation. Neither AI nor crowd-sourced models can replace the judgment, context-awareness, and cultural competency that professionals bring to the job.
- Do not integrate AI generation and content modification capabilities into advertising systems.
- Label all AI-generated and AI-modified content. Do not allow users to remove content provenance information from election-related content.
- Publicly and repeatedly acknowledge the threats AI poses to voters and election officials. Include this warning in all AI products.
- Actively support election officials. Provide cybersecurity resources, multi-factor authentication, and AI literacy and anti-phishing training to election officials, and help them build rapid-response infrastructure.
- Discourage the use of consumer platforms for internal election administration communications. Election officials should not coordinate sensitive matters over social media.
Some in the industry may view these as burdens. They are investments. User trust, once lost, cannot be recovered via press release. People are deeply skeptical of both AI and the companies creating it: Pew reports that in 2025, 50% of Americans are concerned about increased use of AI in daily life, up from 37% in 2021,[8] and our own polling indicates 68% of Hispanic Americans, 62% of Asian Americans, and 60% of Black Americans agree that AI is a threat to democracy.[9] The companies that take these concerns seriously and make AI as safe as possible will earn loyalty that no marketing budget can buy.
Another point to consider is that AI regulation is coming: 80% of Americans support rules for AI safety and data security, even if it means developing AI capabilities at a slower rate.[10] Very few issues in this country are as bipartisanly popular as protecting people from harmful AI. When that regulation comes, companies that are already investing in AI safety and responsibility will be far ahead of those that are rushing to comply. It should also be said that despite the current administration’s stance, disparate impact liability — the legal concept that entities can be held responsible for unintentional discrimination against marginalized groups — remains the law of the land. Future administrations may not be as welcoming to those who violate it.[11]
Both our elections and our faith that our votes matters have long been under attack. Without safeguards, AI is a powerful tool to further those attacks. It is time for AI companies to be accountable for their role in our democracy, the foundation for their business and its success. An environment in which voters no longer believe their ballots count is not an environment under which innovation can thrive. This moment calls for leadership — and history will remember who stepped up, and who stepped aside.
[1] Nathaniel Gleicher, “Coordinated Inauthentic Behavior,” Meta Newsroom (July 8, 2020), https://about.fb.com/news/2020/07/removing-political-coordinated-inauthentic-behavior/.
[2] Vanessa Pappas, “Combating Misinformation and Election Interference on TikTok,” TikTok Newsroom (Aug. 5, 2020), https://newsroom.tiktok.com/combating-misinformation-and-election-interference-on-tiktok.
[3] X, “Expanding our Politics to Further Protect the Civic Conversation,” X Blog (Sept. 10, 2020) https://blog.x.com/en_us/topics/company/2020/civic-integrity-policy-update.
[4] Center for Countering Digital Hate, “Rated Not Helpful: How X’s Community Notes System Falls Short on Misleading Election Claims,” (Oct. 30, 2024), https://counterhate.com/research/rated-not-helpful-x-community-notes/.
[5] Greg Druck, “AI Now Writes as Many Online Articles as Humans,” Graphite.io (May 15, 2026), https://graphite.io/five-percent/ai-now-writes-as-many-online-articles-as-humans-do
[6] Adi Levi, Or Levia, Sardhendu Mishra, and Jonathan Morra, “AI vs. Human Moderators: A Comparative Evaluation of Multimodal LLMs in Content Moderation for Brand Safety,” ArXiv (Aug. 7, 2025), https://arxiv.org/html/2508.05527v1#S5/.
[7] Gabriel Nicholas and Aliya Bhatia, “The Dire Defect of ‘Multilingual’ AI Content Moderation.” WIRED (May 23, 2023), https://www.wired.com/story/content-moderation-language-artificial-intelligence/.
[8] Brian Kennedy, Eileen Yam, Emma Kikuchi, Isabelle Pula, and Javier Fuentes, “How Americans View AI and its Impact on People and Society,” Pew Research (Sept. 17, 2025), https://www.pewresearch.org/science/2025/09/17/how-americans-view-ai-and-its-impact-on-people-and-society/.
[9] Cornell Belcher, “Leadership Conference Civil Rights Monitor,” The Leadership Conference on Civil and Human Rights (Nov. 2025), https://civilrights.org/edfund/resource/2025-civil-rights-monitor/.
[10] Benedict Vigers and Justin Lall, “Americans Prioritize AI Safety and Data Security,” Gallup (Sept. 17, 2025), https://news.gallup.com/poll/694685/americans-prioritize-safety-data-security.aspx.
[11] Chiraag Baines, “When Machines Discriminate: The Critical Role of Disparate Impact Liability in AI Accountability,” The Center for Civil Rights and Technology at The Leadership Conference for Civil and Human Rights (Jan. 22, 2026), https://civilrights.org/disparate-impact-ai/.
Alejandra Montoya-Boyer is the Vice President of the Center for Civil Rights and Technology, where she focuses on research, policy solutions, and advocacy that responds to the growing opportunities and challenges from AI and emerging technologies and their impact on marginalized communities. Previously, Alejandra was the Director of Policy at Prosperity Now and has also worked at the National Association of Counties (NACo), the Bill & Melinda Gates Foundation, the Joint Center for Political and Economic Studies, and the Office of Ben Ray Luján (NM-03). She has a diverse background in policy, political campaigning, grassroots advocacy, and program management with deep content expertise in workforce development, tech policy, economic justice, and racial equity.