AI Meets the US Midterm Elections

Good morning!

Headed into the US midterm elections season, the information environment continues to evolve. Millions of people now use AI chatbots, both to retrieve information about elections and to create it. Last week, the Brennan Center for Justice put out an expert brief tackling the question, “Does AI Fight or Fuel Election Disinformation?” For the podcast, I spoke to one of its authors, Larry Norden, vice president of the Elections and Government Program at the Brennan Center.

Our conversation touched on a range of issues, including how AI chatbots are addressing election-related conspiracy theories and misinformation, how AI is being used to generate false election-related content, the federal rollback of election-security support, and the escalating violent threats, abuse, and harassment faced by election officials and workers. The Brennan Center researchers found that AI did push back on conspiracy theories, but also that responses to election related questions frequently contained inaccurate or false information, and of course AI systems are more than useful for producing such false information, even as they fail to identify it.

One thing we discussed, in particular, is the information void created by the Trump administration, which has removed various publications related to election security, and the problem of what to do if the administration pushes false election claims through official channels. For instance, Norden told me that some of the citation links to election information that chatbots provided in response to their research prompts came back broken:

“A lot of the broken links that the chatbots came back with were government sources. They were Department of Justice, they were the Cybersecurity and Infrastructure Security Agency,” said Norden. “And the reason that they were broken was because this administration took them down. And so I don't know what happens—and I think the AI companies need to be planning for this—when you have formerly trusted agencies like CISA or the Department of Justice putting up bad information. Or do the chatbots continue to draw from them? And there is a controversy then. I don't know the answer to that.”

You can listen to the episode here.

In a related podcast this week, Laura Karpas looked into one new technique that campaigns may try this cycle—focus groups among panels of synthetic voters. She spoke to one researcher who has considered the question of when and how such simulated evidence can be used in campaign research, and another who wonders what the implications may be for the complex interaction between political leaders and public opinion.

Listen here.

There’s a great deal more to tackle this week! Keep reading for highlights from across the contributor community.

Mark Zuckerberg’s ‘path to a positive AI future’

This week, Meta founder and CEO Mark Zuckerberg published an essay called “The Future is for Everyone: The Path to a Positive AI Future.” Zuckerberg says we are “fortunate” to live in this moment at the dawn of the age of AI before staking out the core question, as he sees it. “The defining questions of our age are who will have access to superintelligence and what will we direct it towards,” he writes. “Will it be centralized and restricted to a few institutions, or will it be a tool that empowers everyone?”

• Roy L. Austin, Jr., Meta's former VP of Civil Rights and now director of the Howard Law Artificial Intelligence Initiative, read the new manifestoagainst Meta's record—and finds a widening gap between the words meant to inspire and the company's actions. “Sadly, Zuckerberg’s manifesto is replete with inconsistencies between the words he puts forward to inspire and the actions that he and other tech leaders keep taking,” he writes, “If Meta truly wants to make AI for everyone, its leaders should stop talking to AI and start listening to people who want fewer data centers and better relationships with human beings.”
• Equal access to a personal AI agent isn't equal power, writes University of Oslo professor Petter Bae Brandtzaeg. Personalization "can create an illusion of decentralization"—power appears handed to individuals while staying concentrated in the infrastructure their agents depend on.

When AI safety falls short

• Georgetown's Mark MacCarthy argues the Trump administration's new voluntary AI safety review shouldn't exempt open weight models. Their ordinary use, he writes, poses the same significant, foreseeable risks as closed models.
• Anna Neumann, PhD candidate at the Compliant and Accountable Systems Group at the Research Center Trustworthy AI, Holli Sargeant, a research fellow in law at St John's College, University of Cambridge, and Jat Singh, a professor at the Research Centre for Trustworthy AI, University Duisburg-Essen and University of Cambridge, argue that system prompts alone don't predict model behavior, so AI safety assessments must evaluate a system's outputs, not solely its written instructions.
• Aspen Policy Academy Science and Technology Policy Fellow Venita Subramanian argues that state disclosure laws for AI companion chatbots are a meaningful first step, but that hourly reminders can't prevent emotional overreliance or protect minors in crisis, and that UX design must do more of that work.

On platform accountability

• Courts are finally starting to treat feeds and chatbots as products and to address harmful design, writes Courtney Radsch, director of the Center for Media & Digital Governance at Open Markets Institute and a board member at Tech Policy Press—a turn with consequences for Meta, TikTok, and the AI companies now facing wrongful-death suits.
• "An accessible tool does not have to be an intrusive one," writes Ariana Aboulafia, who leads the Disability Rights in Technology Policy project at the Center for Democracy & Technology. She argues that when Meta introduced its smart glasses, it used disability rights as a Trojan horse for facial recognition that threatens the privacy of people with and without disabilities.
• Brazil has suspended Discord's livestreaming feature in the first enforcement test of its child protection law, the ECA Digital, reports Tech Policy Press fellow Tatiana Dias.

AI, hype and the governance gap

• Nathan Sanders, a data scientist and past fellow at the Berkman Klein Center for Internet & Society at Harvard University, and Bruce Schneier, a current fellow at the Berkman Klein Center, say we must distinguish AI's technological problems from its capitalism problems. Developers are solving the first, they argue, while leaving the second to market incentives. Structural reforms will be necessary to steer industry toward the public interest.
• Governments from the EU to Malaysia and India can pursue very different policies—all branded "AI sovereignty." AI policy researcher Haakon Huynh argues the concept is so fuzzy it can justify almost anything, and governments are spending without defining it.
• Why do we fall for hype again and again? As part of an ongoing series on Hype Studies, educator and organizer Max Haiven makes the case that hype touches "our estranged power as part of a cooperative, storytelling species"—and preys on it.
• Based on interviews with public defenders, Princeton's Dominik Stammbach & Peter Henderson mapped where AI helps—surfacing briefs, making sense of evidence—and where it doesn't. They make the case for careful adoption.

The eye of the state

• India's takedown order to GitHub over the BitChat app used by protestors is "a natural, pragmatic progression" of the state's systematic use of coercive power online, writes Amber Sinha, contributing editor at Tech Policy Press.
• Subhashis Banerjee from Ashoka University examines the evidence behind Aadhaar's core technical and fiscal claims and finds it weaker than its reputation suggests — from self-reported deduplication accuracy to widely cited savings figures.
• With school attendance in Brazilian schools verified via facial recognition serving as a criterion for granting social benefits, the system poses a high risk to children’s rights and the livelihood of thousands of families, writes André Ramiro, researcher at the Chair of Informational and Communication Law at Hamburg University. Now, the Brazilian Data Protection Agency has a say.

Scam compounds

• Sanctions from OFAC and the EU, FBI deployments, ASEAN declarations — and still the scam compounds move deeper into Myanmar and adopt AI tools, writes Laura Scherling, a lecturer at Columbia University. A 2026 OHCHR report finds them "still in place," more entrenched than ever.

What we’re watching

The fight over data centers is increasingly at the fore of American politics. Jill Lepore has an essay on the subject in the Financial Times this weekend, and Sen. Bernie Sanders (I-VT) discusses it with The New Yorker’s David Remnickon The New Yorker Radio Hour. We’ve been covering this issue closely from an international perspective. We invite analysis on this and other AI infrastructure-related matters.

I wish you the best for the week ahead!

-Justin