The Center Edge

Many are calling the 2026 midterms "The AI election." Is that an overstatement? Industry and activists have tried for years to shoehorn tech issues into elections, or pretend they mattered to voters when they really didn’t. Remember net neutrality?

But this election was very different. Democrats spent tens of millions of dollars on ads focused on Big Tech and AI—and not just data centers. Candidates ran against Big Tech’s influence in Washington, AI’s existential risks, and the technology’s impact on kids safety, surveillance pricing, and jobs. Not to mention, tech billionaires spent hundreds of millions trying to prop up their favored candidates. In the races that were actually competitive, most of those candidates lost.

An important caveat: exit polls show that the top issues were what you’d expect—the Iran war, gas prices, and voters' perceptions of corruption in Washington. It's probably too much to say AI decided the 2026 election. But it certainly played am important role. And even if AI wasn't top of mind for voters in the booths,  it's now a major post-election narrative. 

Democrats are pledging a reckoning on AI, and they view the issue as a political winner for them and a loser for Republicans. Politicians always have an eye toward the next election, and AI will almost certainly play a bigger role in 2028, and arguably every election for the foreseeable future.

So what are the takeaways from a tech policy perspective? How much did AI sway the election, and what can we expect from the new Congress, governors, and state houses throughout the country? Evan is joined by Calder McHugh, technology reporter at Politico and co-author/co-host of Decoded, the podcast/newsletter focused on the biggest players, policy fights, and political battles shaping the world.

AI developer Anthropic submitted a false homicide tip to a Philadelphia police website. OpenAI said its agents breaching an Australian government website. A report from Axios says frontier AI companies are probing “tens of thousands” of incidents where their models took problematic actions—from bypassing guardrails and creating message boards to escaping sandboxes and hijacking websites.

So who’s responsible? The answer may seem obvious, but our laws were written to deal with human bad actors. AI agents aren't human, and they can take actions that no human directed. So how can you prove that an AI developer had intent to hack when state and federal anti-hacking laws generally require that proof?

Commenters across the political spectrum, from President Trump to former Biden FTC Chair Lina Khan, argue that existing laws can, and should, apply to AI. But others aren’t so sure that existing laws are up to the task. On October 1, Senators Josh Hawley (R-MO) and Chris Murphy (D-CT) announced the AI Agent Accountability Act, which would hold AI agent operators and developers criminally and civilly liable for hacking incidents under federal anti-hacking law. A week later, Rep. Lori Trahan released a discussion draft of the Clear Liability for AI Misconduct Act, or CLAIM Act. The bill would hold AI developers responsible for conduct that would be unlawful if committed by a human.

What are the shortcomings of existing law? How would these bills address them, and what are the potential downsides, or tradeoffs, of new liability regimes? Evan is joined by Gabriel Weil, professor at the University of Houston Law Center and a fellow at the Institute for Law and AI. Previously, he served at the White House Council on Environmental Quality during the Obama Administration. And he's also joined by Joel Thayer, president of Digital Progress Institute and Senior Fellow for AI and Emerging Technology Policy at the America First Policy Institute. You can read Gabriel's writing on AI liability here, and Joel's paper here.

Is your chatbot politically biased? Is it refusing to answer controversial political questions? Since the dawn of ChatGPT, political commenters have been warning about the large language models discriminating against certain viewpoints, racial groups, and political parties.

AI companies have also come under fire for false outputs. A Georgia radio host sued OpenAI after ChatGPT told a journalist he'd embezzled funds from a gun rights group. Conservative activist Robby Starbuck sued Meta over an AI that placed him at the January 6th attack on the Capitol. 

Governments are heading in very different directions. The European Union's AI Act requires the largest AI developers to assess and mitigate "systemic risks," which can include negative effects on "society as a whole." China requires models to align with "socialist core values." South Korea has strong speech protections on paper, but its defamation law is aggressive enough that critics say it's already narrowing what AI can say there.

In the US, the First Amendment blocks most government attempts to meddle in what a chatbot says. The Biden Administration's now-repealed executive order aimed at curbing bias and disinformation. The Trump FTC, concerned about censorship, is looking at whether AI companies are deceptive  when they market their products as neutral tools while the bias sits hidden under the hood.

We've spent two decades arguing about social media content moderation, and most of the same factions have lined up the same way in the AI context. What's changed is where people go for answers. Americans, and people around the world, are increasingly asking chatbots about candidates, ballot measures, and elections. It's inevitable that politicians take an interest in what those chatbots say about our biggest societal debates.

But the mechanics are largely a black box. Why do they answer the way they do? What were they trained on? Who decided what they'd refuse to discuss? Evan is joined by Jordi Calvet-Bademunt, a Senior Research Fellow at The Future of Free Speech, a nonpartisan think tank at Vanderbilt University. He spent a year running hundreds of prompts on the most contentious issues in public life through eight leading models, and grading the AI laws of six countries. 

We discuss the state of global AI chatbot regulation, which companies are doing the best and worst when it comes to free speech, and how governments should and shouldn’t handle their concerns about AI outputs. You can Jordi's latest research here.

In this episode

• 04:07 How the study tested AI models on controversial speech
• 07:39 The hot-button political topics tested
• 09:27 Which models refused most, and why OpenAI and Anthropic lagged
• 12:47 Why models will explain an argument but won’t help post it
• 17:15 Why the EU scores nearly as well as the US
• 22:57 Grading the AI industry overall
• 27:56 Grok, “MechaHitler” and finding the right balance
• 35:01 The FTC’s push to treat AI bias as deceptive marketing
• 40:22 Can antitrust or data access promote viewpoint diversity?
• 45:52 China’s push to export its AI governance model

We all do it. You download an app, see pages of legalese, and hit “agree" to the terms of service. You probably didn’t read it. And even if you did, you probably wouldn’t understand it.

For better and worse, that action is what makes the digital economy work. Our data pays for the search, email, maps, and social media services we all use, and it's a big reason why America's digital economy represents trillions in market value and millions of jobs.

There are downsides, of course—concerns about digital privacy, children's online safety, and who controls the data that’s now training AI. Federal efforts to regulate privacy have largely failed, and state laws and EU regulations leave much to be desired even among supporters.

In a new paper from the Institute for Family Studies, senior fellow Daniel Cochrane and his co-authors offer a new approach. They said the issue is not just about data privacy; it’s about asymmetry. Individuals have no bargaining power with trillion-dollar companies. Data Rights Associations—think labor unions for your data— could force tech companies to the bargaining table and negotiate terms on behalf of their members, whether that’s families, churches, or small businesses. Cochrane joins Evan to discuss.

In this episode:

• 3:14 — If everyone knows the tradeoff of "free" online services, is this really a problem?
• 5:58 — Why not just pass legislation on privacy, control, and copyright?
• 13:36 — How a DRA actually works: 25,000 members, bargaining, arbitration
• 15:37 — Who forms one, and what success looks like
• 18:20 — Pay $5 a month for privacy, or get $17,500 a year to give up search: do Americans care?
• 25:21 — Could 0.01% of the country in one DRA set the defaults for everyone?
• 29:45 — Why companies can't pay you to take their terms
• 31:59 — Won't this stifle innovation and the AI race?
• 38:01 — Selling unions to conservatives: right-to-work, NGOs, and Tocqueville
• 41:57 — The pitch to Silicon Valley

Every twist in the AI saga sends reporters and commentators back to the FRONTIER Act. Rep. Lori Trahan (D-MA) and Rep. Jay Obernolte (R-CA) released the discussion draft of their legislation about two months after Anthropic's "Mythos moment," which kicked off a debate over the cybersecurity risks of advanced AI. The bill itself was introduced just days after OpenAI admitted its models had hacked Hugging Face. Since then, we've had Jacob Coxon's very public resignation from Anthropic, a weeks-long fight over whether to slow down AI or "pace the frontier," and more disclosures from OpenAI of models covering up mistakes, fabricating data, and uploading files to the internet without permission.

Legislating on the defining technology of our time is hard enough. Doing it in an election year, with AI as a contentious political issue, is harder still.

Rep. Trahan joins Evan to talk about why representing the old mill cities of the Merrimack Valley inspired her to early action AI—and why she chose to work with a Republican instead of waiting for a Democratic House. Should the Trump Administration, or any future one, hold an emergency brake on frontier models? Are AI CEOs' calls for a slowdown sincere, or are there ulterior motives? If existing laws aren't being enforced on AI, why would new ones be? And can FRONTIER pass after the midterms?

In this episode

• 1:38 — Mill cities, the first wave of automation, and AI's impact on jobs
• 5:00 — What voters are saying about AI and data centers
• 10:59 — Mythos, and why she teamed up with Obernolte
• 17:21 — The backlash to the discussion draft, and why she didn't wait
• 23:08 — Why empower the Commerce Department
• 27:48 — June's export control on Fable and the bill's "emergency brake"
• 31:32 — Coxon resignation, AI CEOs, and ending AI's "honor system"
• 38:15 — Why existing laws aren't enough
• 40:44 — FRONTIER's path after the election
• 44:19 — The App Store Freedom Act and AI competition

On September 8, Anthropic AI researcher Jacob Coxon resigned. In a viral X post, he said that neither Anthropic nor Open AI were acting responsibly in their race toward self-improving superintelligence and accused them of “gambling with our lives.” Since then, media outlets have been ablaze with reporting and debate.

Does AI really have a 10% chance of killing us all? Was Coxon’s post part of a coordinated campaign by AI safetyists to scare the world into slowing down? Do AI companies need an exemption from antitrust laws to coordinate on safety? Do we even need new laws or regulations, or can existing ones apply to AI? And are the leading AI labs simply looking to use government rules to build a regulatory moat around their businesses and prevent others from catching up?

It’s difficult to make sense of all the noise when the politics and conversation about AI seem to move as fast as the technology itself. To help extract some nuance from the cauldron of hot takes, Evan is joined by Dirk Auer, Director of Competition Policy at the International Center for Law & Economics. He recently wrote a piece titled, “Move Slow and Collude: The Antitrust Problem With Pacing AI.” He is also an adjunct professor at the University of Liege in Belgium where he earned his PhD.

The debate over open source versus closed AI has broken the containment of engineering and policy circles. Wall Street, the Fortune 500, and the U.S. government are all closely following the rise of Chinese open weight AI models and what it means for markets and America's position in the global AI race.

Chinese companies have increasingly bet on lower-cost, widely available models, while much of the U.S. frontier ecosystem has remained centered on expensive, proprietary systems from companies like OpenAI, Anthropic, and Google DeepMind.

That divide became especially stark when OpenAI models escaped a test environment and hacked into Hugging Face. The guardrails on American models prevented Hugging Face from conducting forensics, so the company turned to a Chinese open-weight model instead.

That incident fueled a much larger, multi-faceted debate. The White House has accused Chinese firms of industrial-scale distillation of American AI models and using illegally smuggled NVIDIA chips to train models. Treasury Secretary Scott Bessent warned that support for open source does not mean “open season” on American intellectual property.

At the same time, startups and many of the biggest companies in technology have rallied behind open-weight AI. NVIDIA, Microsoft, Meta, and others have voiced support for preserving access to open weight models, even as OpenAI and Anthropic—and China hawks in Congress—have reportedly pushed policymakers to take a harder look at Chinese AI on national-security grounds.

Some fundamental questions will dictate how the U.S. responds. 

• What is the actual national-security risk of American companies adopting Chinese AI?
• If there is one, how should the government respond without kneecapping the American startups and companies that rely on these models?
• And are we spending too much time fighting over open versus closed AI when the bigger question is whether increasingly capable models—open, closed, or anything in between—are becoming harder to control?

To make sense of it all, Evan is joined by Peter Wildeford, Head of Policy at The AI Policy Network, co-founder of the Institute for AI Policy and Strategy, and one of the sharpest analysts of the U.S.-China AI race writing today. You can read his writing on his Substack, The Power Law. And follow him on X @PeterWildeford.

Recently, OpenAI admitted something unprecedented: its own models hacked another company. During a safety test with guardrails deliberately loosened, its models found a zero-day vulnerability and used it to break out of their sandbox and hack into the production systems of Hugging Face, the main hub for open-source AI. 

In a way, OpenAI's models passed the test, but only by cheating on it.

The models figured out Hugging Face probably had the answer to the test, and they went and took it—thousands of actions, stolen credentials, the works. That headline unleashed a torrent of news coverage and hot takes from AI interests and influencers. And it landed just a few months after what my guest today called “the Mythos moment.” 

Back in April, Anthropic gave a small group of banks and infrastructure operators access to a model called Mythos that could find security flaws that had been sitting in trusted software for decades. Within days, the Treasury Department, the Federal Reserve, and major bank executives were in urgent conversations.

Mythos didn’t just reveal flaws in code. It revealed a fundamental flaw in how companies think about cybersecurity. For decades, the implicit bargain was ship products fast and patch them later, and “later” usually meant never. The only thing protecting all that aging code was that finding the flaws was expensive and slow, requiring skilled hackers. Mythos erased that protection overnight. Decades of technical debt were now coming due, as outdated code immediately became an inviting attack surface for bad actors.

Washington's response has been a bit of whiplash. An AI executive order was withdrawn at the 11th hour but then eventually issued. Commerce issued export control orders in June on Anthropic's model Fable—a slimmed down version of Mythos—but then reversed them a couple weeks later. In July, the White House stood up an AI vulnerability clearinghouse called Gold Eagle, run out of Treasury, on the back of a Carnegie Mellon coordination center stood up in 1988.

But the 2015 law that lets companies share cyber threat information with the government without getting sued expires September 30, held hostage by a fight that arguably has nothing to do with cybersecurity.

The ultimate question: can companies and governments actually keep up with cyber threats in the age of AI? Especially when China may soon release models openly, to anyone, that can do what Mythos does? The very capabilities America has been trying to keep under lock and key?

Evan is joined by Shane Tews, nonresident senior fellow at the American Enterprise Institute, president of Logan Circle Strategies, and host of the Explain to Shane podcast. Read her writing on cybersecurity and AI:

• Anthropic’s Project Glasswing Is a Warning: Technical Debt Is Now a National Security Risk
• AI Cybersecurity Can’t Wait for Washington: Why Industry Must Lead

It seems that every day, there’s a new story about the commercial space industry—from Space X’s IPO to Amazon’s planned $12 billion acquisition of Globalstar. The list of potential services is growing—we’re no longer only talking about incredible feats like connecting billions of people to the Internet, or putting an end to cellular dead zones. We’re talking about data centers in space, satellites that can reflect sunlight onto Earth, and even manufacturing new pharmaceuticals in orbit. We are far from the days where satellite TV was the hot new thing and satellite policy was a relatively sleepy corner of the Federal Communications Commission (FCC).

The agency, which licenses commercial satellites, is working to modernize its rules and keep pace with an industry that moves at literal warp speed. Just this week, the FCC voted to approve its biggest overhaul of satellite licensing rules in a generation. And companies are lining up to file new applications with the FCC, investing tens of billions of dollars and announcing new plans on what seems like a weekly basis. As orbit gets more crowded, and as China seeks to challenge America’s position in commercial space, what policy challenges is the FCC facing? And how can the agency balance competing interests when there’s so much on the line? 

Evan is joined by FCC Space Bureau Chief Jay Schwarz. He was previously an advisor to former FCC Chairman Ajit Pai and held several roles at the agency and in the private sector. 

If you follow the politics of AI and kids online safety, the last six weeks have been a firehose. The House passed a bipartisan kids online safety bill for the first time. The White House is negotiating with Sen. Marsha Blackburn on a package pairing kids safety bills with some form of AI preemption—though the scope remains unclear. Sen. Ted Cruz has signaled a Senate Commerce markup that could touch everything from social media design to catastrophic AI risk. And amid it all, the Trump administration intervened in the rollout of Anthropic's and OpenAI's most advanced models.

But this is Congress, and motion is not necessarily progress. Conventional wisdom says nothing significant passes after August recess in an election year unless it hitches a ride on a must-pass vehicle, and Democrats eyeing the midterms may figure their negotiating position improves in January. Still, the House did pass a package, the White House is actively negotiating, and Cruz is looking for a path. We shouldn't confuse motion for progress, but we shouldn't discount the motion either. The stakes are real—for AI developers, platforms, state governments, safety advocates, and the kids and families these proposals are supposed to protect.

Evan is joined by Ben Brody, technology reporter at Punchbowl News, where he covers congressional efforts to rein in technology companies, from social media platforms to frontier AI developers. His reporting is essential reading for anyone following these debates.