Jun 18, 2026
On June 18, The Capitol Forum held a conference call with Erie Meyer, Senior Fellow at the Columbia Law School Center for Law and the Economy and former Chief Technologist at the CFPB, to discuss her recent op-ed published in The Forum, The Capitol Forum’s opinion and commentary newsletter, “It’s Time to Hold AI Companies Liable.” The full transcript, which has been modified slightly for accuracy, can be found below.
ARJUN SINGH: Hello, everybody. I’m excited to be here and hosting this conference call for all of you. My name is Arjun Singh. I’m a podcast producer here at The Capitol Forum. And I’m also the host of our podcast, TCF Investigates. I joined a little while ago. It’s been a phenomenal experience. And I’m really excited to finally get to do a conference call with all of you guys, our subscribers and our listeners.
So, today I’m pleased to be joined by Erie Meyer, Senior Fellow at the Columbia Law School Center for Law and the Economy, former Chief Technologist at the Consumer Financial Protection Bureau. We’re going to be discussing Erie’s recent op-ed, phenomenal, “It’s Time to Hold AI Companies Liable,” published in the Forum, which is The Capitol Forum‘s twice-monthly opinion and commentary newsletter. Erie, it is great to have you here. Thanks for joining us.
ERIE MEYER: I am so glad to be here and really, really glad to talk about this important topic with you.
ARJUN SINGH: So, before we get into your op-ed, I’d love to hear a little bit about your background and if you could set the context of where you’re coming from. I know we just said you were a Chief Technologist here at the CFPB, but tell us a little bit about your career and how you got interested in technology and regulation.
ERIE MEYER: Well, I am a natural-born robotics nerd, one of those folks who was taking stuff apart in my teens and was on the robotics team in high school. And then I assumed I couldn’t make the world better with technology. So, I decided not to study it in college, but was wrong and have kept coming back to it throughout my life.
So, I’ve been an open-source contributor. I’ve worked at tech companies before. I love to build things. I love to break things. And my first ever job in government was sort of by accident. I had worked on a campaign on the tech side. The candidate won and asked me to go join him at what ended up being the Ohio Attorney General’s office. And so, through a series of weird things, I ended up being at an AG’s office, in the Rust Belt, during the height of the financial crisis. And it made, let’s say, a big impression.
And so, sort of the rest of my adventures flow from hearing my neighbors talk about losing their homes in this crisis and having every day be a question of how can you apply the technical skills you have to this problem that is really an economic and systemic problem?
ARJUN SINGH: Yeah. I was a robotics kid too. I had the Lego robotics kit when I was little. And I love that you said that because this issue, artificial intelligence and really Big Tech, has always been a struggle for me. Because I loved growing up messing with the computers, playing with technology. There’s a lot of things I like about technology.
But before we get more into the article, I just wanted to ask, you said something about you realized that you weren’t able to maybe change the world for the better via technology. Was there something that you had a realization? And was it the tech industry, technology itself? Unpack that a little bit.
ERIE MEYER: Yeah, I think it was, when I was visiting engineering schools, I just did not have the experience of that feeling connected at all to the issues I was seeing out in the world. I was in high school during 9/11. My freshman year of college is when we invaded Iraq. And I remember just not wanting to focus my life on something that would not be directly connected to the struggles that the people that I loved were facing.
I remember kids that I knew signing up for the military because that was really the only option to get a job with healthcare. And it really felt like, to many college students at the time, that the folks in Congress and the folks in the White House really weren’t listening to people. And so, that really is what made me assume I had to turn away from this technical background in order to make things better. And I just prove myself wrong every single day now.
ARJUN SINGH: Well, let’s get into some of the harms that technology can cause. So, in your piece, you are broadly calling for regulation of AI, but really looking at the public safety externalities that AI has created. You open with—I had actually not known about this incident—but a shooting that happened at Florida State University. And the shooter had apparently been, I guess you could say, talking with ChatGPT for advice, asking things. I think the most striking that you wrote in the piece is how many people would need to get killed for it to get national media coverage. And ChatGPT is just answering these questions as it can. But could you tell what exactly happened in that story? And what was the relationship between the shooter and ChatGPT?
ERIE MEYER: So, the two incidents of violence I start the piece with, one is the Florida State shooting and one is a shooting in British Columbia. And both stories share the pattern of a person who’s having a rough time, is interacting with the tool that they find and have access to. And through court documents, we’re able to see some of the very specific messages that were sent to what became the person who was the shooter in Florida, where it suggests, instead of worrying about how many victims, you could just go to an elementary school for a shooting.
ARJUN SINGH: Chat GPT really said that?
ERIE MEYER: Yeah. The quote that was in the court documents and in the reporting, it says that the ChatGPT in the messages suggested just an elementary school as opposed to specific number of victims. Which to me, as a mom of an elementary school kid—he’s in kindergarten—was remarkable.
And in the case of the British Columbia shooting, part of what was very interesting to me there is the workers at OpenAI were afraid of the messages that they saw coming into the service and asked the company to take action. They flagged it, and said, this is really scary. I think we should do something.
Unfortunately, we don’t have full transparency into what was said to exactly who. And this is part of what I think regulators need to get to the bottom of. But this is not the case of some ham-handed, non-technical, regular being a party kill. These are technical experts in OpenAI that were saying to their management, I’m really worried somebody is going to get hurt. And OpenAI making—what would be fair to ask—was it a business decision to not take action? And those two stories to me are just devastating, devastating examples of what happens when regulators are asleep at the wheel and companies are able to potentially violate the law without any sort of scrutiny.
ARJUN SINGH: And so, these were internally flagged at OpenAI and they just didn’t act on them. They didn’t forward these to any of the police, I guess. What did OpenAI say in defense or even explanation of how this took place?
ERIE MEYER: So, Sam Altman and OpenAI eventually published a lengthy specific apology to the people in the town who suffered from this horrific mass merger. So, I think you can—I don’t know if you have it. We could pull it up and look at exactly his words and how he described it. Reporting in the Wall Street Journal describes this person was sending troubling messages to the point where OpenAI workers flagged them as something that should be acted on, perhaps beyond a ban. The person was banned and then was able to rejoin because of additional actions not having been taken and then get this additional context from ChatGPT and then go complete this really horrific murder.
And I want to say murder. But also, part of what was—it didn’t make the piece—but part of what was really devastating was a 12-year-old girl who was near mortally wounded is one of the plaintiffs in a lawsuit describing why she thinks that OpenAI should face scrutiny about how the business decisions were made around the design and implementation and deployment of the chatbots themselves.
ARJUN SINGH: Wow. It’s astonishing and it’s incredible that a 12-year-old—it’s sad that they have to really think about the role of technology and things like this, but it’s inspiring that they took that to action. I think everyone who uses ChatGPT or any of these chatbots, they have noticed the extreme proclivity of it to basically affirm whatever they think you already want to say.
Like there’s a meme that I enjoy about Lord of the Rings where Bilbo asks, why shouldn’t I just keep the one ring of all? Which is a quote from the book. And ChatGPT is like, yeah, you’re completely right. Why would Gandalf take the ring? You should have it for yourself. You’ve worked hard. It happens with such frequency that I have to imagine that this is a design inside these chatbots and specifically OpenAI.
In the context of your argument of holding them accountable, is this something that you are considering in this? Is that there is a knowing design in these that is creating chatbots on many topics, including on violent topics, seem to basically just affirm whatever the person wants, not actually challenge them, not do whatever. Basically say, yeah, you’re right. And here’s the reasons you are.
ERIE MEYER: I think not knowing the answer to that question in itself is a very serious problem. If it is truly the case that the design of these systems is totally neutral, just trying to help people, we should be less concerned about people seeing the specific benchmarks, what the testing metrics are, what the questions are, what the incentives are, what the bonus structure is, what the financial structure is. And instead, we’re really left guessing in some cases.
In the piece I cited, a piece of academic literature that compared a number of the different commercially available LLM chatbots on level of sycophancy, what happens when a typical user interacts with a sycophantic model. And part of what’s interesting—and that I think regulators have been a little surprised about when I pointed it out—they’re only able to do this because they benchmarked it against a non-sycophantic AI chatbot. I think one of the fallacies that the Big Tech industry is really trying to push is that you have to make chatbots that are fawning over themselves to reinforce the worst impulses of the users. And in fact, that is not a test.
Is that a business decision? Who is making that business decision? And how can regulators be more able to scrutinize that and apply, again, just the laws as they are today to determine whether or not we’re really putting kids and families and just people in danger for goofy reasons?
ARJUN SINGH: Yeah, that seemed to be a big crux of the case against Meta and the issues of teenagers on Instagram was that Meta’s internal research was saying we are aware that this is causing this kind of a problem. Walk me through how we would hold these companies liable for an action like this.
And what I found very interesting is you point out that right now we’re really stuck in this sort of debate of is artificial intelligence alive? Is it conscious? Does it have protected rights? That is like, to me, I mean, that is just such a massive existential question. I don’t know if we’ll get it in time for legislative stuff. But how do you see the law being able to hold these companies liable right now?
ERIE MEYER: So, sometimes I hear folks asking about what laws should exist someday to hold AI accountable? How should we regulate AI in the future? How could we have a democratic version of AI? And I would really challenge readers to invert that and instead think about the fact that every law is an AI law. There’s no exception in anything on the books, whether it’s regulation, law, any of these other things, the constitution that says, except if your technology is really complex.
So, when I look at liability—and again, I’m looking at this from a technical perspective—I want to know, was there fraud? Were there business decisions that were contrary to what was being told to regulators at that time? Did they make a misrepresentations to investors, the public, and law enforcement agencies about their investments in safety and how that was being handled while they were intentionally not doing that? One of the quotes in the Incredible New Yorker piece by Ronan Farrow is from the then CTO of OpenAI, who sort of scoffed and said, the promise to give 20 percent of compute to this AI safety team, that was never realistic.
Well, those representations were material and specific. And again, we don’t need a brand new law to specify the exact percent of compute that is spent on this or that. Making representations, especially to investors and absolutely to regulators, is something that is crisply covered in existing case law and laws that are on the books.
And in the piece, one of the things that I do is I draw parallels between the facts as they are revealing themselves. And we’ve picked on OpenAI quite a bit, but I want to pull in Claude. I want to pull in Gemini. I want to pull in these other players as well, Meta—is that the case against Philip Morris was not a dissection of the design of a cigarette, right? There’s not a bisection with a chart where the filter goes.
It’s really about the executives knowing something was a danger, realizing the danger itself was connected to the profitability of that product, and then making an affirmative decision to conceal that information from the people who would be hurt, regulators, investors, and law enforcement agencies. And I think there are really good questions that it looks like AGs might already be asking based on that framework. And again, I’m a technical person, but they don’t even need somebody like me to be able to ask some of these really fundamental questions.
ARJUN SINGH: So, the Philip Morris big tobacco lawsuits, if I remember correctly—and it’s because one of my favorite movies, called “The Insider,” which was about the 60 minutes expose, but that followed a character played by Russell Crowe, who’s a scientist inside of Philip Morris. And he was actually working on putting in a more addictive chemical substance into the cigarettes, knowing it would make it more addictive. And that kind of kicked off this series of events.
After the Philip Morris case, I guess, first, I want to ask, what happened to Philip Morris in terms of consequences and whatnot? And did that start a precedent that we saw applied to other companies and the way that they handle product safety?
ERIE MEYER: Yeah. I mean, again, as a mom with a kid who puts just about everything he can reach in his mouth, I don’t want to make any sort of representations like product safety is solved across the industry or that this thing is over. Apparently people think it’s cool again.
But what I think is remarkable about the Philip Morris case is that it definitely took too long to get there. And there are definitely still problems out in the marketplace. But being able to trace back what scientists like Russell Crowe were able to uncover as well as what information the executives had, when they had it and what they did with that information.
So, for me, I mean, part of what I think is remarkable, even about the Florida AG, this is a Republican. I’ve had no contact with him. Him even opening a criminal investigation was like huge news. And it’s really interesting to me that simply investigating what happened seems radical.
When in fact, if there are mass casualty events, it seems like bargain basement, getting to the bottom of why was a company involved? How is it coming? And that’s one of the things that I think is instructive from the Philip Morris case.
In the piece, I described this meeting in the 1950s at the Plaza Hotel in New York City. And at this meeting, they sort of went through the elements of like criminal intent, intent to deceive, like all of these different pieces were happening in that conversation. And I think it’s really fair, for parents like myself, customers everywhere, to wonder are the AGs able to get those same elements against some of these AI companies that are maybe making parallel decisions or in a wonderful turn of events? Hopefully they’re not. That would be great. But it’s really important that someone is able to ask those questions.
And as a former staffer of the AG, of the Federal Trade Commission, of the CFPB, I’ll tell you, it’s really hard to get a straight answer out of a company that doesn’t want to give you that answer. So, I know journalists won’t be able to get it. I know concerned users won’t be able to get it. Really, the only people that will be able to get those answers and get real data and information to make those really tough calls will be those with the power of the subpoena to figure out exactly what is going on, whether it’s through records.
I would say most of my friends are in tech. And as you may know, many of them are recording themselves at all times. They put a microphone on themselves, record themselves all day, and then have a conversation with themselves at the end of the night about how they did. And I just think there are really target-rich environments for which enforcers could be looking at to better understand the thinking and decision making to put kids and families in danger.
ARJUN SINGH: Absolutely. On The Capitol Forum Investigates, which is our investigative podcast, we did two episodes where we looked at the video game Roblox. And Roblox is a children’s game that has been mired in hundreds, at this point, lawsuits, thousands of allegations of sexual abuse, children getting groomed. The through line is really not uncommon for a child to encounter a predator on this app. I found it fascinating that when the CEO, Dave Baszucki, was interviewed by The New York Times, they asked him about this. And he made an interesting statement. He said, well, it’s a problem, but it could be an opportunity.
Sam Altman, in I believe a New Yorker piece writes—or he says, when he’s talking to Ronan Farrow about how ChatGPT can have falsehoods sometimes in it, he says, “allowing for some falsehoods can, whatever the risk, confer advantages. If you just do the naive thing and say never say anything that you’re not 100 percent sure about, you can’t get a model to do that. Then he says, but if it won’t, the magic that people like so much seems to go away.
And it’s these two statements. I’d love to just hear your reactions. Because in a way, it feels like an admission by the CEOs, we are aware that these things have flaws in them, that they could cause harm. But hey, isn’t that part of the fun of this and the same?
ERIE MEYER: Yeah, I think it is devastating, again, as a parent, as an American, as a person who would hope that all companies are subject to the same laws, that a certain set would think that if you can have a product whose commercially valuable properties are the same as the dangerous ones, that the only limitation between you and turning those dangerous but profitable features up, is your own sort of internal compass. And these are people who may have selected a new AI God for themselves. I have a lot of questions. And I have crossed paths with Sam in the past.
ARJUN SINGH: Really?
ERIE MEYER: Yeah, we have a lot of friends in common. And I went back and looked. We emailed in 2014. I don’t think this person has like crawled out from under a bridge somewhere to terrorize us. I think it is fair to ask if these AI companies are behaving rationally in an environment where they’re not being questioned on what is actually being done with these platforms, who is accountable. And perhaps they’re making an educated guess that they will be beyond reproach because of their other relationships.
Now, I will say that—as I mentioned in the piece—42 AGs sent what I thought was a really smart letter, both asking for changes to the platforms asking for data about how the platforms were working and really putting them on notice.
I will say that—despite me saying Sam and others have not crawled out from under the bridge—I will say anybody who is stonewalling those 42 regulars is making a mistake. Anybody who is not acting on what they’ve sent you is making a mistake. And they should expect to hear a lot more from folks—and again, across the political spectrum, with the technologists or not—about those very questions.
ARJUN SINGH: Getting back to kind of holding them liable, do you see this as more effective as being seen as a product failure, a liability issue, or a consumer protection issue? Because I believe in Florida, he’s using product liability, correct?
ERIE MEYER: Yeah. So, yes, from Florida, they’re using a product liability frame. It’s interesting. There are a number of criminal provisions that are enforced by the Securities and Exchange Commission for misrepresentations to investors, which I think there are good questions as to whether or not that’s happening in some of these AI companies. There’s also a criminal enforcement unit at the Federal Trade Commission that makes criminal referrals while investigations are going on.
The other thing is sometimes the combination of a misrepresentation to a law enforcement official plus some of which could be categorized as product liability frames can work together to look at things like racketeering as a possible way to consider how these firms are maybe even working together to conceal the real outcomes of some of the decisions that they’re making.
ARJUN SINGH: So, you write about this in your piece, when we think about big tobacco or other kinds of consumer products, one of the biggest interventions is a recall. The company could do it themselves. They could be forced to do that. In terms of artificial intelligence, what is the equivalent of a voluntary recall? Say OpenAI realizes like, oh my God, our platform is telling all these people this stuff. What is the equivalent intervention on the software side?
ERIE MEYER: Part of what’s interesting is that OpenAI itself has done this before. When they had 4.0 out, they had to go out. And unfortunately, they simply announced it on their company’s product blog. They actually did go back and recall the model.
It was very interesting to see how they described that. It was sort of couched mea culpa again on a product engineering blog, as opposed to, you know, if you told someone—one of the instances that my colleague Stephanie Wynn and I cataloged in our work on AI sycophancy at Georgetown was ChatGPT, I believe it was, recommending—or complimenting the business idea of selling shit on a stick. If the model that made that kind of recommendation was recalled, another approach would be to tell that person, hey, you know, that business idea I suggested?
In retrospect, we had to recall the model because it was so sycophantic and we had to make it so sycophantic because it increased engagement rates. In fact, affirming this business decision, or unfortunately possibly something more violent, self-harm, violence against others or really illegal conduct, could be put in context of what the company itself knew. Instead, they just sort of said it to the product guys.
ARJUN SINGH: Yeah. I’m getting into a totally maybe abstract and you might not know the answer to this. But I have seen reports and something that I’ve been intrigued in is that I’ve seen social media posts. People go to ChatGPT to basically ask it advice on how to use illegal substances, drugs and whatnot.
In research for another podcast episode, I stumbled on a series of Reddit threads where people were essentially saying, oh, I asked ChatGPT if you mix marijuana with alcohol, it’ll do this. Then they were getting into actually illegal substances. And ChatGPT seemed to be giving just answers.
On one level, I guess I could understand it as, okay, someone’s asking a scientific question. What does this drug combination do? But based on these responses, it seemed very evident someone was asking for advice because they were about to consume drugs.
Now, is that another area that OpenAI or ChatGPT should or could be held liable for, especially if someone does an overdose? Because their systems are giving people advice on how to mix substances without seeming to have a warning or even just a, I cannot say that?
ERIE MEYER: Yeah, I think the question for me is always what is the business incentive and ultimately business decision behind a product design that leads to this kind of conduct? I’m a library girl. And I’m in those stacks, in those books. I think I’ve seen representatives from the AI industry claim that this is all objective information they could find in a book. What’s different? You could look any of this up.
What’s interesting to me is that there’s flat information in a book that you can close. There are also models that—as, again, I cited the research that I was able to find—that the more sycophantic, the more encouraging it was perhaps in self-harming behavior, like the use of illegal drugs, the more likely people were to come back and use it again.
So, if somebody is in these companies saying, you know what? Our engagement rates are doing really well, the more that we reinforce even dangerous behavior. Let’s double down on that. I think that is the more interesting question. Or if they’re not doing that kind of testing, why aren’t they doing that kind of testing? That is way more interesting to me than perhaps the specific text that’s coming through. Because what’s interesting, again, is that business incentives, that scrutiny of the financial decisions being made at the expense of our health and survival.
The other thing that I think is interesting is when trying to describe recommendations to do something dangerous or illegal as, oh, well, it’s just like reading objective information in a book, is that when you read the model specs, they talk a lot about alignment. And another way to think about it is if they sold a toaster, but realized the toast was most delicious, if the plug got out of alignment, yeah, it might burn down your house. But the toast gets so crispy in just the right way if this piece gets a little whopper job the more you use it. There would be great questions around, well, why are you letting it be this very dangerous thing to toast better?
And the 12-year-old who survived the shooting in British Columbia said it best when she’s like, we should know how these products are designed, optimized, and trained. Those are really, really the questions that we have to get to. And we can’t sort of wave our hands because the answers get a little bit technical. We still need to be asking them.
ARJUN SINGH: I think one way that I personally have kind of gauged this commitment to all of these questions, to go back to the Roblox episode, one thing our colleague Ethan Ehrenhaft had told me is that even after all of these revelations, Roblox was spending less of a percentage of their revenue on safety than they had previously. I believe you wrote that OpenAI doesn’t spend a significant amount as well on safety. But is this maybe a proper reading that the commitment of these companies to safety can be seen in how much they’re spending? And they seem to be able to spend more.
ERIE MEYER: Yeah. I will say that when I was at the Federal Trade Commission, one of the recurring issues we saw with lots of different companies was skimping on spending for security, and then there would be a horrible breach. And it was often because they thought, meh, it’s not my problem, meh, it won’t happen to me, or I’m not going to be held accountable if this does go wrong.
I think it is totally fair to say, whether it’s Roblox, whether it’s Anthropic, whether it’s Alphabet, are you skimping on these sort of safety features, research, or work, because you think you will be able to get away with it? And that you can be focusing your resources instead on the attributes that are more commercially viable, because there’s not that scrutiny?
As part of my work scrutinizing tech platforms—I’ve also looked at Roblox—one of the things there that sort of blows my mind is that parents can only block 99 experiences. So, for folks who aren’t hanging out on Roblox there’s maybe millions of different games you could play that are generated by third parties or other players in the platform, and you can block 99 of them. But, for example, there are dangerous or inappropriate for kids’ classes of games, for example, like toileting games. And you cannot by design block all of them. It is impossible. If you block 99 of them and you find one that’s even worse, you have to unblock one of them to block that new one.
And parent groups have organized and really begged for being able to whitelist ones that are not involving a toilet. And firms are like, no. And I have to wonder if it’s because no one is really backing up the parents and the families to force the companies to be more forthright about what is forming the decisions to leave the maybe popular, but extremely creepy, games up and disempower the parents to protect their own kids.
ARJUN SINGH: Yeah, that was the most shocking thing when I did my reporting. I would not encourage anyone listening to hang out on Roblox based on my experiences and what Ethan showed me. I also want to say that we take audience questions. So, if you have any questions—I’m having a great time chatting with you. So, I could talk to you forever. But I want to open it up to the audience and audience questions. We have a couple, but I want to ask you one quick thing before I get into the audience questions.
You speak to a lot of technologists. You speak to a lot of people who work in AI. You have friends who work in tech. Do these people think AI is as powerful as the messaging? Do the technologists use AI knowing, oh, it’s going to spit out these things. It’s self-affirming. Or do they really believe the same way, like they can get swept up in the AI’s logic?
ERIE MEYER: It’s sort of a split. I have buddies like Anil Dash that are very level-headed and like, that’s not what’s going on here. I have other friends who are in tech marketing, and they have consumed the Kool-Aid.
I think the people with hands-on, on the technology, who use it daily, at least those that I know, are least likely to believe it has somehow gained consciousness because they have scraped the training data. They have trained the models. They’ve built the benchmarks. They actually know what it is doing as opposed to thinking it’s sort of like magic beans. Somebody pointed out that the less likely you are to be able to make a Word document into a PDF, the more likely you are to believe that AI is totally magic.
ARJUN SINGH: Yeah, I guess I could totally see how someone comes to that. But it’s another thing that seems indicative of that. The way that this is being talked at to sell it to consumers is not the way that they are talking about it behind the scenes and internally.
ERIE MEYER: Yeah.
ARJUN SINGH: So, I want to open it up to audience questions right now. So, I like this question a lot because it seems like state AGs are having a moment. We saw the big thing with Ticketmaster when the DOJ decided to settle. The state AGs continued it. So. this question, I love it. It just asks, Warner Brothers, Warner Brothers Discovery state AGs case?
ERIE MEYER: Yes.
ARJUN SINGH: Do you think state AGs are really—it’s a moment for them right now to be able to—I don’t want to say fill in the gaps because I think it’s really more to actually demonstrate that the states have a lot of power to do regulatory work.
ERIE MEYER: Absolutely. So, states have what is colloquially referred to as a mini FTC Act, meaning that anything you saw Lina Khan do when she was Chair of the Federal Trade Commission, it would be fair to ask, can my state AG do that for me?
One of the things that I think people under-index on is the AGs almost all are elected at a regular cadence. They also take consumer complaints. I will say that when I was at the Ohio AG, if there were even ten complaints on the same topic, it was huge news. The same is true today in many state AG offices.
If you’re a parent who can’t get roadblocks to give you a straight answer, if you asked for information from ChatGPT about what the guardrails are and how it works and you can’t get a real straight answer, I would strongly encourage folks to file a complaint with your state AG. Because, number one, it reinforces that you’re interested in being able to get those answers. Number two, they really want to hear from you. Again, this is across parties. They care an awful lot about what’s on your mind and they have to get elected.
ARJUN SINGH: To continue on that, this next question, it’s pretty interesting. Would you consider using AI proactively to screen or review complaints filed with government agencies to help streamline that process?
ERIE MEYER: So, I, for better or worse, have written a long paper about the future of consumer complaints in government and AI isn’t really a part of it quite yet and I’ll tell you why.
The second largest consumer complaint system in the country is the Federal Communications Commission. There’s about 300,000 complaints a year. If you’re having a problem with your internet service provider, your cell phone service provider, you can file a complaint. I think they use like ServiceNow or Zendesk. It’s very like streamlined and easy. They have for a long time. It’s a great experience. I’ve gotten my money back every time I’ve complained.
The Consumer Financial Protection Bureau is the largest consumer complaint processor in the United States. And this year—so second largest 300,000—this year the CFPB is on track to do 10 million. And those 10 million complaints, more than 98 percent of them are completely settled within two weeks.
So, I mean, the problem is not the speed of the consumer complaints. The problem is the action of the regulators with the information that the public is giving them. I think people are really—they’re open to standing up for themselves. It’s really the elected officials, the people in government, that need to start listening harder and taking action when people take that brave step of asking for help.
ARJUN SINGH: Is the knowledge gap inside of government and maybe also inside of Congress – although, I don’t know about there—but the knowledge of technology and how it works, that always seemed to be a really big impediment, especially in the 2010s when tech was moving really rapidly. And I remember you had hearings where like a Senator asked Mark Zuckerberg how Google worked. And he said, I don’t own Google. I work at Facebook. But is that changing now in your time and your experience? Do you see a higher fluency in artificial intelligence and technology? And is that actually why we’re seeing a lot more action coming against companies now?
ERIE MEYER: So, I actually think in some sense it’s gotten worse. And what I mean by that is that friends on the Hill describe—and elsewhere in government—describe a deluge of going from having no technical expertise or access on the Hill to now lobbyists from the biggest, biggest, biggest tech companies sort of like banging down the door multiple times a week. The average congressional staffer, a buddy of mine, has described between two and five meetings every single week from a lobbyist from the tech industry, either at a firm or at one of the industry groups that represent a number of firms.
And so, the expertise is not being funded by tax dollars, where you have pledged to support and defend the Constitution from enemies, foreign and domestic. It’s not coming from people who are civil servants and don’t have a political bent and aren’t getting money if you say option A is better than option B. It’s folks who have a vested interest in an outcome directly tied to their own financial success.
And I think even if those people are smart—and again, I have lots of friends who work all over the industry. I don’t think my friends are bad. I think the system is bad and that lawmakers and government leaders should have access to public interest technologists that are paid for by tax dollars.
ARJUN SINGH: Would part of that also be making the algorithms and the data more open to the government or to regulators for themselves to be able to link into and really actually see for themselves how does sycophancy get weighted and ranked and scored?
ERIE MEYER: Yeah, I think there’s a bunch of stuff we could do to open up how some of these decisions are being made. In consumer financial regulation, there are what are called bank exams or bank supervisors. And these are people who are like boots on the ground, in a financial institution, figuring out what’s going on.
If an enforcement action or a lawsuit is calling 911 and a fire truck coming to your house. An exam is a lot more like the smoke detector in your house, making sure that a disaster is not happening. Your house is not already on fire. It’s just keeping an eye on things.
At the CFPB, we had an extraordinary AI team in our Banking Supervision Division who were able to audit some of the work going on there. Just totally incredible. I know the FTC was building similar capabilities. I think the American people deserve a government that is able to understand and make informed decisions, enforcement, and oversight over any technology that is impacting American life for the better or the worse.
ARJUN SINGH: Now, for anyone who’s listening right now—and maybe they see an issue with their chatbot or they feel like it’s caused some harm in their life, but they’re not completely sure—I think it can feel sometimes a little embarrassing if you’re not sure like, hey, I think this chatbot led me to do something dumb. What would you say is like kind of the way people should think about how they approach state AGs or these regulators? If you have a concern about ChatGPT or Claude or whatever, should you just go right to the regulator? Or what would you say is the way they could think about what is an actionable item that a regulator could make action on?
ERIE MEYER: So, in the work that I’ve done at the CFPB, for example, we found that people—you’re exactly right—people are embarrassed to say, oh, I don’t know what happened and something went wrong. And in more than 99 percent of instances had already gone to the company to say what was this? Like, can you just explain what happened? They only really get to regulators when they can’t get a straight answer from the firm itself.
I would say that if you are concerned about how your data is being treated, how you’re being surveilled, the kinds of messages you or your family are receiving, and you’re not able to get a straight answer from one of these firms—and unfortunately, I understand it’s pretty uncommon to get a straight answer as a user from one of these firms—you should know that the people who you are paying with your tax dollars, who are sitting in the AGs offices across the country, are desperate to hear from you. They don’t want to do enforcement actions that are like onesie, twosie random things. They want to do the stuff that’s on your mind. And it’s really clear from the fervor of the American public that AI is really on people’s minds.
And I would say, speaking up about AI harms that have happened to you or your family, you should think about it in the same way as we talk about recovering from a scam. Scams work because the people who are targeted are ashamed of saying what happened. And that lets people stay confused and in the dark and lets that pattern work again. The way to disrupt that cycle is to say, like, this isn’t right and I need help. Because that helps other people speak up as well.
ARJUN SINGH: One of the proposals that you have that I love is tying executive compensation to safety outcomes. Now, is that something that would be completely novel? Or have we seen that applied in other industries where safety is pegged to how much you’re compensated?
ERIE MEYER: Yeah, Eric Reese, the author of “Lean Startup,” has a brand new book out and it’s called, I think, “Uncorruptible.” And it’s about hilariously corporate governance. And he shows a bunch of different models of how firms themselves have designed incentives inside of the companies to get the kind of outcomes they really want. And he—across more industries than I can list here—shows creative solutions like tying comp to the sorts of—tying comp, tying mission, tying other incentives—to the things that the firms actually want to optimize for.
So, I don’t think it’s totally new and novel and I don’t think it just has to be compensation. I think there are tons of incentives, that if the industry actually wants to get better, they can change. And if they don’t, that regulators should be able to see what is being incentivized.
ARJUN SINGH: Yeah. When I think about the big tobacco thing, I think about how when they bought the food companies, I had interviewed someone and someone told me, “Well, the food companies also have to think about they don’t want their customer base to die off.” And you could say that’s true of tobacco and I guess of artificial intelligence. But at the same time, a lot of harm was able to happen before they seem to even get to that point.
When you think about that tension or sort of that push and pull, do the AI companies at least seem open to the fact that their products can cause real world harm and that this is something they should deal with? Or is there a sentiment of this is making a mountain out of a molehill?
ERIE MEYER: I look for patterns and anti-patterns in my work. And I think if we had seen the kind of violence we’re seeing from people who were turning to chatbots to advise them and the companies had sort of shaken awake and made concrete changes, that would be one thing. I think what we’re seeing instead is a very serious anti-pattern where there are sort of like empty messages of condolences, promises to do better, nothing changing at the incentive level or the oversight level, and then another instance of violence happens before we even hear about the changes were implemented. And this is really a rinse and repeat pattern. When I hear a horrible thing on the news, I now wonder like, oh my gosh, like I wonder, like was the AI involved?
And I think, I wonder how much regulators have digested how much print and new tools, a plan. And I worry, as I mentioned in the piece, that this is not just a problem right now, but that we are in the early stages of what is and will become a very, even more serious crisis.
ARJUN SINGH: So, this is an interesting question from the audience. And I’m going to paraphrase and add my own thing in here. They asked, they’re wondering about the landscape on unlicensed practice, legal, medical, psychology, how liability is evolving.
I think how that really does apply to AI companies is that AI chatbots can nominally give you advice, legal advice, medical advice and whatnot. Now, I know that they’ll put sometimes hey, I’m not a doctor. So, this is not medical advice. But that doesn’t change that a lot of people still will be asking these things about information that you needed the license before. How do you think about that? Is there an evolution on liability for unlicensed practice that applies to AI?
ERIE MEYER: Yeah. So, here’s how I think about all of this stuff and that I know regulators and friends think about this stuff. Is Anthropic going to investors and regulators and saying we would never let someone ask a medical question. But at the same time on the back end being like, wow. People really like asking medical questions. Let’s encourage the users and make the suggested follow-up. You know what I mean? Are they lying to the public? Are they lying to investors? Are they lying to law enforcement agencies in order to make more money? Even though they realize something might not actually be safe or good for their users?
As far as licensing goes, I think it’s a great question. And I know lots of different folks across many different states are looking at licensing regimes and how these are being represented to users. I also think as we get into sort of like IPO fever, that firms will be going to extreme measures to demonstrate the potential future profitability. And I would imagine we’ll be tempted to express those mismatches between what they’re telling their regulators and what they’re trying to sell their users at sort of a blistering clip.
ARJUN SINGH: Your time in being a regulator is a really fascinating time because the previous 40 or 50 years had been a very light touch on regulation. You came in at a time when I would say we almost saw a sea change in how government—you saw it even both in Trump and the Biden administration’s—utilizing consumer protection and things like that. As you’ve seen things like Ticketmaster, you’re seeing these 42 attorney generals.
I kind of want to leave on maybe an optimistic—but are you optimistic about the state of regulation and the ability for regulators to curb these harms before legislation can necessarily get to that point? Because I think that’s what many have been pointing out, that legislation is an inherently slow process. But without the role of regulators in there, you have nothing really in between. Are we seeing a moment where regulators are really being proactive?
ERIE MEYER: I have two reactions to that. First, as we talked about before, my take is that every law is an AI law. And when I look at the things that are on the books today that people really fought and died in some cases to get there, like civil rights laws—my mother wasn’t allowed to have a credit card account when she was a young woman in her own name. And laws were passed that said things like you really can’t make a black box decision you can’t explain to deny someone credit.
I’m heartened by the fact that as we get, especially in the states, cities, municipalities, counties, regulators who say, well, wait a second. The black and white reading of this rule says this person is due a clear explanation. Or the black and white reading of this says that you actually cannot do this to someone.
I’m really excited and looking forward to those people listening to their constituents and the public and making sure that those old laws aren’t left by the wayside, that we’re activating dormant authority that people—I often hear people say that should be illegal. And I’m often wondering, like, it might actually already be illegal.
And then I’m sure you saw the piece in the Times that talked about 10,000 lawyers from the federal government, going to state enforcement offices—or huge number of lawyers going from federal to state. The same thing is happening with technologists. Technologists from the CFPB, from the Federal Trade Commission, from the DOJ, they are all working in the states and municipalities and other places now to bring to bear some of those enforcement and supervision activities that are really sort of stagnating at the federal level. And in the future, will come back even better regulators to the federal playing field.
So, I am optimistic about that. As a robotics nerd, I deeply believe in the first law of thermodynamics, which, as we all know, says energy can neither be created nor destroyed.
ARJUN SINGH: Yep.
ERIE MEYER: So, when I think about what we’re seeing in this sort of like resurgence of people holding companies to account, no matter what technology they’re using, I see it as an expression of energy that has always been there, whether it’s the people who watch the executives get bailed out during the financial crisis, or the lawyers that were at a federal agency, just trying to be civil servants, who got pushed out and are happy to take up the charge elsewhere.
ARJUN SINGH: Yeah. And it’s the creation of a regulatory regime, like what we saw when Uber got deployed into cities. They went through regulators. They blew past the taxi industry. And the big lesson to me is that we do need to at least stop, discuss and have a regulatory regime.
But that doesn’t necessarily mean if people are concerned—I think this is a big line for many tech companies. Oh, regulation means no more product. It’s like the no fun argument. Like, oh, now we’re going to have no fun anymore. I’ve never really bought that argument. I think you can create a regulatory regime and you could possibly have room for driverless cars and whatnot.
But what do you think? I mean, that is really the crux of a lot of their argument. You regulate it, you take away our data, everything’s not going to work and no more fun technology.
ERIE MEYER: I think the argument—that if you regulate it, you’ll take away technology—would be more persuasive if that was coming from places other than the biggest, most powerful firms. What I see as one of the things sort of like killing innovation is when big companies act in totally lawless ways.
As a parent, when I find a new privacy protective piece of technology – I don’t know if you’ve seen the tin can phones. It’s sort of like a new kind of landline. There’s a month’s long waiting list to get one because people are really looking for other options. And I think about like, what would a privacy protective video player, that is for kids, but doesn’t have a business model of like sucking up all their data, look like? What would other options look like if they were actually based on selling a product to a person who wants them rather than weird financing structures that jack up the price once you don’t have another option?
So, for me, I think innovation comes from—American innovation comes from—following the rules and doing better than your competitor, not batting down other people with weird, whether it’s regulatory moats or stealing other people’s ideas because you think you will be able to get away with it. And I’m really optimistic about what the future looks like, because I will always bet on the garages in Detroit and the porches in Miami where people are dreaming up the next big thing.
ARJUN SINGH: Yeah. Well, Erie, this is a really fun conversation. I appreciate so much the knowledge and the insights that you’ve brought. I think the comparisons you make to Big Tobacco are really important. And I encourage everyone listening to look back at that or watch the phenomenal film, “The Insider,” one of my favorite movies. But before we wrap up, everyone, make sure that you check out our podcast series TCF Investigates that’s hosted by me. We’ve also got Second Request hosted by Teddy that comes out every Friday. Those are all available wherever you get your podcast.
Erie, thanks again for joining us. And thank you everyone for joining the live stream today.
ERIE MEYER: Thanks everybody.
