Why candidates need to hear your AI questions now
What happened when I asked congressional candidates about AI risk
Guest post by Haven Harms
Meet the Candidates
I walked into the assembly hall at Crown Point Circle, my sweaty hands clutching the small binder that held my printed-out question. I claimed an aisle seat a few rows from the front for easy access to the microphone, and my husband sat down next to me. Candidates Heidi Hall and Chris Bennett settled into oversized armchairs onstage. The moderator announced that Congressman Ami Bera would try to call in from D.C., but might not make it. That evening in early March, I was at the Meet the Candidates event hosted by the Democratic Women of Nevada County to hear from the candidates running to represent California’s 3rd Congressional District.
Unlike many focused on AI governance, I don’t live in San Francisco or D.C., but in the rural foothills of the Western Sierra, in a town that makes you think, “Wow, this looks straight out of a Hallmark Channel movie!” (it actually is the setting for the 2006 Hallmark Channel movie, The Christmas Card). For the past five years, I have had one foot in this town, building a life, and the other immersed in a community of professionals concerned about where AI is headed. Congress will be making consequential decisions about AI in the coming years, and the people casting the votes are sitting in armchairs at forums like this one, answering questions from whoever shows up.
I showed up. My heart pounded as it all became real: I was going to get up, speak into the microphone, and ask these candidates about AI.
A constituent, not an expert
Questions are one of the most useful tools in a functioning democracy: they reveal truths about candidates and voters alike. When candidates hear their constituents asking about AI, they can’t say, “No one cares about AI risk.” I vote in every election, but I also show I care about AI risk through the questions I ask before I do.
When the moderator opened the meeting up for questions from the audience, I wanted to be strategic about timing. My question is kind of intense, I’ll wait a bit so it’s not so jarring, I thought. Admittedly, this was also to give myself a chance to build up the internal tension that would propel me off my seat like Squints from the movie The Sandlot declaring “I CAN’T TAKE THIS NO MORE!” and jumping into the deep end of the pool despite not knowing how to swim, so that Wendy Peffercorn would have to save him. (If you know, you know.)
The first audience member, an older woman with graying hair, praised the candidates’ earlier talking points, and then asked, “How do you feel about AI, and do you think it would be important to regulate this industry?”
Wait, what?
My husband and I whipped around to look at each other, jaws dropped. This was an assist I did not expect.
Notice what she hadn’t done: cited a study, named a bill, used fancy jargon. It wasn’t a technical question, but it effectively got everyone on the same page about where the candidates generally stood on AI regulation. This woman established that I wouldn’t be weird for asking about AI, making my question all the more askable. She was there as a concerned constituent, and that was enough to put AI on the agenda for the evening.
A position, or the absence of one
When you publicly ask candidates a more specific question, it forces them to take a position—or exposes that they don’t have one. Either way, you learn something a campaign mailer would never tell you. A dodge in front of a full room carries a cost, and gives candidates a reason to have a better answer next time.
It was time: off my seat, to the microphone, no turning back. I began setting the scene: “This past Friday, the Pentagon ended its military contract with Anthropic, one of the leading AI companies, and designated Anthropic a supply chain risk—a label typically reserved for foreign adversaries—because the company insisted on restrictions against the government using their AI for mass domestic surveillance and fully autonomous weapons.” Despite the glare for taking up time from the woman in charge of the microphone, I continued, “So my question is, what role do you believe Congress should play in setting legislation for the government’s use of AI? And are there lines around AI usage that you believe the government should not cross?”
The first audience question concerned AI regulation generally; mine concerned the government’s own use of AI. I anchored it to an event widely covered just days earlier and ended with a clear ask: Which lines would you draw?
Bennett, who has a tech and military background, named two real limits: AI shouldn’t be part of decisions that need a human being in them, like decisions about whom to bomb, and “warrantless surveillance” of Americans has to end. Hall, a county supervisor, said she had recently attended state-level workshops on AI and presentations by lobbying organizations. “I had the opportunity to listen to a lobbyist for SEIU who is focusing exclusively on AI, and that was the most frightening presentation I’d heard all day,” she said. Hall argued that the federal government should not prevent states from regulating AI. She praised Anthropic for standing up to the Pentagon and said we should support AI companies that do the same. Then Hall claimed the earlier AI question had already covered mine—it hadn’t—and pivoted to how counties are adapting to AI without naming a single limit on the government’s use of the technology.
I asked my question in good faith. That was precisely what made the redirect costly: when a fair question gets a dodge, the candidate loses control over how the non-answer is interpreted.
The record outlives the campaign
Asking questions at public forums creates a public record on the issue. I can tell you exactly what Bennett and Hall said because the exchange was recorded and posted to YouTube by Nevada County Media. Public records help voters track positions and create an accountability trail for follow-up.
Neither Hall nor Bennett made it out of the primary, but the record outlives the campaign. If Hall were still in the race, I could ask: “In March, you praised Anthropic for drawing lines on the government’s use of their AI model. Do you agree those are lines the government shouldn’t cross, and what would you do in Congress to draw them?” Follow-ups anchored to a candidate’s own words are especially hard to escape: candidates can redirect away from your framing, but it’s much harder to retreat from their own.
A specific question plus a public record can change the outcome of an election. One of the most vivid historical examples comes from the 1858 Illinois Senate race between Abraham Lincoln and Stephen Douglas, champion of popular sovereignty (to let each territory vote on whether to allow slavery). At one debate, Lincoln pressed Douglas to reconcile his doctrine with the Supreme Court’s Dred Scott decision, which had just ruled that territories couldn’t ban slavery. Did Douglas still believe the people of a territory could lawfully exclude it? Douglas answered with a loophole: slavery couldn’t survive without local police regulations, so a territory could exclude it in practice simply by refusing to pass them. The answer played fine in Illinois, where he won the Senate seat. But when Douglas ran for president two years later, the Southern wing of his party read his answer as a betrayal — he’d described, on record, how to nullify their victory in the courts. Rather than unite behind him, the Southern Democrats split off and ran an additional candidate, dividing their vote and handing Lincoln the presidency.
So, you want to ask a candidate about AI…
Campaign season is the best time to ask candidates about AI: they want your vote, so they’re out at town halls and community events, easier to reach than they’ll ever be in office. Here are the principles I’d hand to anyone stepping up to the microphone:
You don’t have to be an expert. Your job is to be a constituent. This isn’t a debate; you’re asking questions during an audition period to see how they will govern. A question from a local voter can carry more weight than a polished one from an out-of-district advocate. You don’t have to be eloquent. The most important part is showing up and asking.
Use broad questions strategically. Asking where a candidate stands on AI regulation in general can introduce the topic and create space for more specific questions.
Build a specific question, and give it a runway. Specific questions block the exits to candidates’ comfort zones. Anchor yours to a major recent event with widespread coverage to establish context and relevance. End with an ask that demands a concrete answer: Name one bill you’d co-sponsor. Identify one use you’d prohibit. Tell us which lines you would draw.
Ask in good faith. You actually want an answer, not a stumble. A “gotcha” buries a trap in the framing, handing the candidate a legitimate exit and letting them play the victim of the question. A fair question is costly to dodge.
Don’t measure success by the quality of the answer. The question succeeds when it leaves your mouth. An answer gives you substance; a dodge still gives you information. This holds even when the format works against you. The next candidate event in my district is a telephone town hall, where you submit your question ahead of time and a staffer decides whether it goes on air. I’m going to ask about AI risk. Even if my question never makes it past the screener, someone on the campaign still heard that constituents care about this issue.
The next election is around the corner. Ask about AI now.
Recently, I’ve been thinking about the groundwork needed for an international agreement to prevent a dangerous race to superintelligence—the kind of deal that becomes real and durable through Congress. My greatest concern is superintelligent AI that could deprive humanity of control over its own future, and potentially cause our extinction. Your concerns might look different—jobs, surveillance, what AI is doing to your kids—but the method is the same.
Not sure where to find an event to ask your question? Check campaign websites, political party calendars, social media pages, and local organizations such as libraries, chambers of commerce, and service clubs. If nothing appears, email the campaign and ask when the candidate will next be available to the public.
The window is now: the general election is Tuesday, November 3, 2026, and the months before it are go-time for asking candidates questions. The people elected in November won’t get to stay stanceless. Congress is weighing whether frontier AI companies should face binding federal safety requirements and whether states are allowed to regulate AI development at all. And as the Pentagon’s fight with Anthropic shows, no one has yet drawn statutory lines around the government’s own use of AI.
Your representative will cast votes on these matters. A question asked now puts the issue on their desk months before it’s on the House floor, giving them a reason to form a view while there’s still time to form a good one.
Start your search. Find one upcoming event for a candidate on your ballot. Think of a question. Then get up to the microphone, and ask.
Haven Harms is the founder and principal of Harms Research & Consulting, supporting organizations working on AI safety, governance, and advocacy.
The analyses and opinions expressed on AI StopWatch reflect the views of the individual contributors and the sources they cover, and should not be taken as official positions of the Machine Intelligence Research Institute.
 | A guest post by
|