Essential AI Literacy for Elections Professionals in 2026

Just a couple months into 2026, elections officials are facing a busy season leading up to the U.S. midterms. Among their many concerns are—or should be!—the impacts of AI on their work, for better and for worse.

AI is a technology capable of empowering both good and bad actors. Beyond actively seeking out constructive uses for AI, the best way elections staff can equip themselves for this new, fast-changing landscape is by getting informed: AI will undoubtedly be a factor in this election season, so it’s essential that administrators learn the terminology and the tools. In addition, creating proactive policies can help elections professionals limit the risks and enjoy the benefits of AI.

Knowing the dangers

At least for this next election, AI is not so much creating entirely new types of threats to the electoral process as magnifying existing ones. Elections staff should be alert to three categories in particular:

1. False information

While misinformation and disinformation are nothing new this election cycle, AI has enabled bad actors to create more convincing photo, audio, and video deepfakes of real people and use them to circulate false ideas. Rapid improvements in the technology in just the last year can produce results capable of fooling even well-informed voters. In 2024, for example, deepfaked videos of Elon Musk appeared to show the tech mogul endorsing President Trump’s reelection as a huge win for cryptocurrency investors—a front for an online scam promoting fake crypto schemes.

Once this content exists, AI can facilitate its spread by instantly generating social media posts or fake news stories and tailoring them to specific audiences and locations. This demo shows how AI can create fake tweets like the one below, which appears to be from a real Virginia news outlet telling would-be voters they have to pay for parking at the polls.

This is a particularly insidious form of voter suppression, because for small towns and cities, it would be difficult to validate. It’s subtle enough news that there might not be much coverage if it were real, and might fail to prompt a correction through official channels.

The spread of misinformation isn’t limited by borders, as we’ve seen in campaigns by international propaganda agents to interfere in U.S. elections. Advances in AI are adding fuel to those efforts. Last year, Anthropic, maker of the AI chatbot Claude, identified an operation that used Claude not only to create false political content, but to manage over 100 bot accounts that disseminated and boosted it to real users online, in multiple countries and languages. It’s an example of AI taking on a more complex role in malicious influence schemes, one usually reserved for humans. AI reduces the barriers to orchestrating similar operations reaching U.S. voters.

2. Attacks on election systems and data

Seasoned elections administrators already know about the possibility of an attack on their office’s IT system, database, or other infrastructure. What’s new is the potential for AI to make these campaigns more sophisticated, more effective, and more automated.

AI turbo-charges old school phishing attacks, collecting information on targets from across the internet and using it to write them highly personalized emails. The enhanced personalization and flawless English in these messages, as shown in this demo, could make it harder to tell that they’re scams and mislead elections staff into sharing sensitive information.

As with propaganda machines, technology companies are also starting to see AI tools used as higher-functioning, decision-making agents in cyberattacks. A November 2025 report from Anthropic describes what it suspects was “the first documented case of a large-scale cyberattack executed without substantial human intervention,” in which AI was used (successfully in some cases) to attempt infiltration of roughly thirty large companies and government agencies. The people running the campaign were able to offload 80 to 90% of their tasks to Claude, including testing for security vulnerabilities, harvesting credentials, and gathering data. This enabled the attacks to be executed in a fraction of the time required for human hackers. Elections infrastructure needs to be hardened to resist this new generation of cyberattacks.

3. Attacks on AI tools themselves

Elections staff are beginning to use AI to streamline their own work, from chatbots that help them write emails to more specifically elections-focused products. While careful use of AI tools for “background” elections tasks can be hugely beneficial, it comes with its own set of risks.

Last year, Microsoft found a flaw in its AI assistant, Copilot, that could have allowed a hacker to get sensitive data from an account simply by sending an email to the owner—no action required by the target. Though there’s no evidence anyone was actually hacked this way prior to the issue being fixed, this kind of “zero-click” attack is a reminder that AI-based technologies may have critical vulnerabilities. These could expose those systems—and any others they’re connected to—to takeover by hostile actors. The nascent security practices of AI tools, paired with the sensitive information they may have access to, make them tempting marks.

This is especially concerning given the number of AI products being marketed to elections professionals, which offer everything from reading registration forms to verifying voter eligibility across record sets. While most of these aids use traditional forms of machine learning, some vendors offer tools that incorporate decision-making AI agents to process paperwork. This kind of technology, especially when it’s used to read documents originating externally, risks what’s called prompt injection—hidden instructions in those documents telling the AI to steal voter information, for example.

General AI guidance for elections offices

As elections teams contemplate whether and how to integrate AI into their work, here are a few overarching rules to keep in mind:

• Provide trustworthy tools staff need. Many people already rely on AI to streamline small steps in their workflows, so consider ways to proactively incorporate it. If the office doesn’t provide a secure tool, employees may use their own, unsecured one.
• Consumer tools are generally not safe enough. The best practice is to get a gov-grade or corporate account for the office, typically through a cloud provider like Microsoft Azure, Amazon Bedrock, or Google Cloud. Gov-grade and corporate accounts tend to have stronger and more stable privacy agreements than their consumer counterparts, and offices with particularly sensitive data may want to set up zero data retention configurations.
• AI security practices are new and still developing. The youth of the industry coupled with the speed at which it’s grown means security protocols at AI companies haven’t had time to mature. Incidents like the hacking of OpenAI and court-ordered data retention should act as reminders that security and privacy policies at these companies are evolving, and their tools should be handled with caution.
• AI should not be trusted blindly. Elections staff should double-check AI outputs before using them to make decisions.
• Be especially careful with agentic AI. Use extreme caution when giving AI tools the ability to access systems and change data—it multiplies the potential harm they can cause.

The midterms are months away, but the advancement of AI capabilities isn’t slowing down. Elections offices should use the time they have now to train staff, vet vendors, and tighten cybersecurity protocols. Those that do will be far better positioned when AI-powered threats arrive alongside the ballots.

This article was co-authored with Arizona State University’s Mechanics of Democracy Laboratory & The Elections Group.