Arizona’s AI election security group pushes for learning lab, more cyber defense

In recognition of the growing threats posed by artificial intelligence, a new security advisory committee in Arizona this week published its first recommendations, including one to create an AI elections lab that can fine tune election-specific AI models for use by election offices.

The 18-page report was published by the state’s Artificial Intelligence and Election Security Advisory Committee, a group recently formed by Arizona Secretary of State Adrian Fontes, who in the report notes his interest both in tackling AI’s risks and harnessing its “transformative potential.”

The committee — which is cochaired by Chris Cummiskey, who runs a cybersecurity and homeland security consulting practice, and Gowri Ramachandran, director of elections and government at the Brennan Center for Justice, the nonprofit law and public policy institute at the New York University School of Law — notes in the report government’s widespread need for increased efficiency.

“Local and state election officials collect and store large amounts of data and are underfunded and under-resourced,” the report reads. “This creates a significant opportunity for AI tools, with proper human oversight, to help election officials continue to administer safe and secure elections in today’s threat environment and efficiently improve voter experiences.”

The committee goes on to recommend the state establish a “learning laboratory, potentially hosted by a major university,” that offers “customized” AI education for election officials. It also urges election officials to craft incident response plans and to continue running table-top exercises, the latter being a preparation that Fontes said was “crucial” in securing the 2024 election and reducing the disruption caused by bomb threats on polling locations in his state.

The report notes that AI can be used by election offices to protect against increasingly sophisticated cyberattacks, and that those cyberattacks, and misinformation campaigns, are likewise fueled by the latest AI models. Arizona is meanwhile challenged by a dearth of data, the committee writes, because private companies often won’t share the data it needs, for privacy or other legal concerns.

“Data access poses a barrier to some efforts to harness the potential of AI to help officials and protect and mitigate against bad actors’ use of AI, which could range from supercharged phishing attempts to corrupting AI-training data to achieve harmful ends,” the report reads. “Access to comprehensive datasets from technology companies would provide policymakers and researchers with the information they need to do their part as we collectively face the rapid integration of AI into our society, including election administration.”

But still, the report points out some current and future uses of AI in securing elections, such as reducing errors in voter rolls, spotting incomplete forms and creating voter outreach materials — all major challenges for “chronically underfunded” election offices, according to the report.

Matthew Jordan, a Pennsylvania State University professor who teaches a course on media and democracy, told StateScoop he agrees that election offices are underfunded, but that embracing the credo of “doing more with less” and infusing more technologies into the democratic process will not help build the public’s trust.

“It’s all about the vague notion that, in the future, look at this cool new thing that’s going to allow us to do that much more. Everything we know about AI so far is that it is riddled with errors, that people don’t trust it — rightfully so,” Jordan said. “Every day on social media somebody does a screengrab of a huge mistake one of these things makes.”

Instead of using technology to compensate for small budgets, Jordan advocated for governments to put more resources into forming nonpartisan groups that carefully deliberate over democratic processes, and meanwhile displaying their work to the public.

“When we want to increase trust in democracy, what you do is put people of all stripes in charge of a process and you let them do what they need to do, and it cultivates that democratic disposition,” he said. “If the problem is that people don’t trust elections, I think we need to slow the process down, put more people in charge of it and let everyone see the work.”

But as the Arizona secretary of state eyes further uses of AI, he already claims success in using the technology to solve several recent challenges faced by the state. Fontes deployed AI chatbots to support staff in responding to hundreds of thousands of “critical” texts and phone calls leading up to last year’s presidential election.

And Fontes’ concerns of AI have are also leading his office to preventative measures, like the creation of a repository of images signed with “content credentials,” designed to boost trust that the images are genuine, a growing concern as political campaigns and bad actors turn to AI to create false images of their opponents and advance false narratives.

“Ultimately, the challenges we identify for election security in an age of AI are all interconnected, and they demand a comprehensive approach that requires collaboration between election officials, technology companies, academics, civil society, the public, and regulators,” the report concludes. “Transparency mechanisms and advanced AI analytical tools can work hand in hand to counteract the evolving risks posed by AI to our democracy.”