Can Democrats Unite on AI Regulation?

In a live interview for The Lever ’s subscribers, David Sirota spoke to New York Assemblymember Alex Bores, the author of New York’s landmark AI safety law . In addition to AI regulation, they discussed the launch of Who Decides, Bores’s new effort to get Democratic presidential candidates on the same page around AI safeguards.

Watch the full discussion:

▶️ Are you subscribed to The Lever ’s YouTube? Click here to make sure .

With the recent Hugging Face hacking incident and a former Anthropic researcher’s doomsday tweets, Sirota and Bores delved into how concerns about AI safety are at an all-time high. They debated where existing law falls short, and how the strongest fix might be personal and financial liability for AI executives. They end on the political reality that AI money is already shaping which Democrats speak up, and how Bores’ new group, Who Decides, is an attempt to get the party aligned on a shared AI agenda before the 2028 elections.

---

TRANSCRIPT

Following is an unedited transcript of this episode. If you plan to quote any part of this transcript, please first confirm that it is correct by listening to the audio.

Host: Hey, everyone. Thanks for tuning in to another episode of Lever Live. Today we're talking about, what else? Artificial intelligence, the fears that it's gonna murder us all, and the concerns that politicians are too corrupt to try to do anything to stop that from happening.

And we are joined by someone who can help us get some informed answers about all of that. We're gonna be joined by Alex Boris. Alex Boris is a Democratic member of the New York State Assembly representing Manhattan's 73rd District. He has a computer science background, worked as an engineer before entering public office. In Albany, Alex Boris became one of the most prominent state-level lawmakers working on frontier AI safety, sponsoring New York's landmark law which requires large AI developers to maintain safety plans and report serious incidents.

This year Alex Boris ran for Congress in a race where AI policy was a big, big deal, and there was a lot of spending by AI-linked groups on both sides, which became a major issue in that race. He has now launched a new organization called Who Decides — an initiative that aims to help Democratic-aligned organizations build some kind of AI policy agenda ahead of the 2028 election.

So we wanted to have Alex Boris on to ask some basic questions. First and foremost: as AI systems become more powerful and more unpredictable, what should the law actually require, and who should get to decide what the law actually is?

So we are joined by Alex Boris. Alex, thank you so much for being with us.

Alex Boris: Thanks for having me. Great to be back.

Host: So I wanna start — before we get to what the law should be changed to — I wanna ask first and foremost: what has changed in the last few weeks about our perception of the threat of AI? What should ordinary people — people who don't know all the details about how AI works, the back end, the experiments — what should they be concerned with? What's real, what's not real, what's at the frontier, what may be on the horizon?

Alex Boris: In some sense, nothing has changed, in that a lot of these threats were predicted. But in the other sense, everything has changed, because we are seeing them now.

So the big news from the summer was OpenAI... an incident where agents hacked this website called Hugging Face, a repository of other models. The agents that they were testing weren't supposed to talk to each other. They ended up communicating and scheming and hacking this website. There were a lot of security flaws in how OpenAI set it up, and ways it could've been done more safely. And also, the agents used a bunch of novel cybersecurity vulnerabilities they figured out — even if it had been perfectly set up, they found ways to elevate privileges and get around some of the guardrails that were there.

That possibility of AI doing that is something that's been predicted, but here it is in the real world, taking an action that if a human had done it would've been a federal felony. I think there are lots of good arguments that it still is a federal felony, despite some vagaries around what it means to have intent in a situation when dealing with AI, but we can come back to that.

Clearly there are real hacks in the real world occurring, and these companies have been talking about the threats they see from AI for a while, and that's weird. I think normal people would say, "Well, if you're so worried about this, why are you building it?" Clearly there's something else going on — lots of conversations about how they're maybe hyping up an IPO, and certainly the marketing departments of these companies are in hype mode, 'cause that's what marketing departments do.

But the Anthropic researcher last week who resigned and put out the statement that no company is approaching this safely — he said a lot of things that have been said before, but in doing so he also gave up his shares at Anthropic. So it was very clearly someone who is not financially incentivized to hype it up, giving the same message, and that then broke through in a way that most of these messages haven't before. And so you're seeing elected officials wake up to the idea that they have to do something. You're seeing people ask what the plan is. And the real question now is: how do we develop a plan that is actually helping people and isn't a giveaway to industry, or isn't just papering over some very real problems that need to be fixed?

Host: Okay. So before we get to what should be done, I just wanna go a little bit deeper on what happened. My question is about how we distinguish between dangerous misalignment and ordinary bad software. One response to all of these examples of AI doing bad things may be that the AI is showing us weak permissions, poor cybersecurity, badly designed software — rather than necessarily an AI system slipping "out of control." So what separates an ordinary engineering failure from evidence of a deeper control problem? What should the public infer, and not infer, from these cases?

Alex Boris: It's a great question, and it's sometimes difficult to draw that line. In fact, Anthropic, in their initial response to a congressional inquiry recently, said everything was just the result of not setting it up correctly, and there's been no misalignment. And then afterwards they're like, "Okay, actually, we're probably wrong about that. We need to specify there is some evidence of misalignment."

Create a FREE account, or log in.

Registered readers of The Lever get access to a limited number of free articles every month, breaking news alerts, and our Lever Daily newsletter.

SIGN UP FREE

Great! Check your inbox and click the link.

Sorry, something went wrong. Please try again.

Already have an account? Sign in