[music]

Brooke Gladstone: When the Pope released his AI encyclical, the co-founder of Anthropic sat beside him and gave a speech.

Chris Olah: How will we ensure that the gains of AI are shared globally? It is an unsolved problem.

Brooke Gladstone: Is it really?

Brian Merchant: We have historically proven tools that are actually pretty good at redistributing wealth and power. They're called taxes.

Brooke Gladstone: How Anthropic markets itself is a moral leader. From WNYC in New York, this is On the Media. I'm Brooke Gladstone.

Micah Loewinger: And I'm Micah Loewinger. Also on this week's show, we visit an Epstein files reading room displaying printouts of 3.5 million files, 17,000 pounds of evidence.

David Garrett: The first printer that we sent this project to came back and said, "We work with a lot of lawyers. We're really worried that if they find out that we printed this, they're not going to work with us anymore."

Micah Loewinger: It's all coming up after this.

[music]

Brooke Gladstone: From WNYC in New York, this is On the Media. I'm Brooke Gladstone.

Micah Loewinger: And I'm Micah Loewinger. On Tuesday of this week, yet another executive order.

News Clip: Trump just signed an executive order this morning centered on government access to advanced artificial intelligence.

News Clip: That lays the foundation out for federal testing of the world's most powerful AI systems before they're publicly released. Now, the testing would rely on voluntary collaboration from the nation's top AI companies.

Micah Loewinger: This, after earlier in the week, one company became the frontrunner in the race to become the AI startup that rules them all.

News Clip: Anthropic has filed what's likely to be one of the largest IPOs in history. By filing, Anthropic has leapfrogged its rival, OpenAI.

News Clip: We reported just last week that Anthropic closed a massive financing. It was $65 billion, $965 billion valuation, almost $1 trillion.

Brian Merchant: Anthropic has built what many in the industry and many users consider the best-in-class AI product.

Micah Loewinger: Brian Merchant is a technology journalist and author of the book and newsletter Blood in the Machine.

Brian Merchant: Their Claude chatbot is one of the favored chatbots anywhere, especially for commercial use, for automating work functions, especially software coding. Whereas OpenAI has been dabbling in consumer chatbots in video generation with its Sora app, that's now recently shuttered. You name it, and OpenAI is in it. Probably key to outmaneuvering OpenAI more recently, Anthropic has really leaned into their reputation as the safer and more ethical AI company.

Micah Loewinger: How did Anthropic get this brand as the ethical AI company?

Brian Merchant: Anthropic began when Dario Amodei, the now CEO and former employee of OpenAI, left the company, citing ethical concerns with Sam Altman's managerial behavior. He leaves and resolves to found a more ethical, a more safety-focused AI lab from the beginning. They are hiring safety researchers, AI ethics people.

Micah Loewinger: They got another bump of press coverage in late February, when Anthropic had a big dispute with the US military. Anthropic had a major contract with the Pentagon, but the US government said they would stop using Anthropic's technology because of the company's red lines, that they wouldn't spy on Americans or let their tech be used in autonomous weapons. Walk us through how that moment was metabolized by the media.

Brian Merchant: The press presents Anthropic as taking the ethical route.

News Clip: Anthropic is imposing another level of morality or control on the use of its technology, which is part of its MO.

News Clip: They are taking the moral high ground here and clearly have a set of red lines for how they feel their technology should or shouldn't be used.

Micah Loewinger: They stood up to Pete Hegseth. [laughs]

Brian Merchant: Exactly, and then they served this absolute gift of Sam Altman running to Hegseth and the Department of War, saying, "Ooh, we'll do it. We'll take this contract where we'll surveil Americans and lend our technology to autonomous killing machines." In truth, yes, maybe there's this red line, but Anthropic is partnering with Palantir. It's lending technology to companies that are building mass surveillance apparatuses already.

Anthropic is working with the government in the first place, with the Department of War. It was involved in providing technology that led to the capture and kidnapping of Maduro and in the bombing campaigns of Iran. That's not hidden information. The fact that they say, "Oh, well, we won't do these things," suddenly, they look like the good guys.

Micah Loewinger: In April, Anthropic announced a new artificial intelligence model that it said was too powerful to be released to the broader public. They started by just offering it to a handful of executives, though they're starting to expand. You say this garnered them reams of credulous press coverage.

Brian Merchant: Yes.

Micah Loewinger: Break it down.

Brian Merchant: To me, this is the most nakedly staged event. This is the clearest example of Anthropic's ethics slop. They make this big show of withholding a model that they say is too dangerous for public consumption, like, "Oh, we've built this model Mythos. It's so powerful that if we gave it to the public, then it could collapse the entire cybersecurity infrastructure of the world's digital systems as we know it. We have to hold on to it. We're doing the ethical thing."

To my surprise at this point, the media did, by and large, buy their line. I'm just going to read a couple of headlines that I pulled from The New York Times here. "You Can't Use This AI. Claude Mythos Preview is Dangerous, Anthropic said. We Explain the Risks." Here's another one, also just from The New York Times, "Anthropic Claims Its New AI Model, Mythos, Is a Cybersecurity 'Reckoning.'" This is actually a classic move that AI companies have made even before the AI boom really heated up.

Micah Loewinger: It sounds so much like Sam Altman saying, "Please regulate us." This technology, if it falls in the wrong hands, will end civilization as we know it.

Brian Merchant: Anthropic gets to have it both ways. It gets to look like it's built this transformative, totally disruptive technology that could change everything and, if we're not careful, destroy it, too. You can just hear the CFOs around the world salivating, "What could it automate if it can do this? What kind of jobs could it replace if it can collapse world infrastructure?" That's the point. They want to feed this continuing hunger for the most powerful AI systems.

Micah Loewinger: Yes, you say that narrative plays a really important role in attracting the interest of VC and raising money. Can you give some other examples of that?

Brian Merchant: The biggest example is the entire concept of AGI, or artificial general intelligence, which was, at least in the early years of the AI boom, the goal of all of the companies to achieve AGI, which is a system that is as smart as a human nominally, but is defined when you look at their charters, as a system that can do most meaningful work of a human worker.

If you're hoping to raise enough capital to compete with one of these giants that are worth over $1 trillion, like Google and Apple and Microsoft, you need a story that's going to generate tens of billions of dollars of investment. You need the biggest story. There's all these factors contributing to the story, not just being, "Oh, we're going to build a product, and we hope people like it. We think they're going to buy it," to "We're going to build the be-all, end-all, automate every job machine," and you want to get in on the ground floor.

Micah Loewinger: Just to point out the irony here. On one hand, Anthropic is raising money on this, "We're the ethical AI company." On the other, they're also benefiting from the hype of the major consequences that would come with realizing their sales pitch, replacing human workers and all of the negative consequences that would have on our society. This contradiction came to a head at the Vatican last month. Anthropic co-founder Chris Olah said this at the press event where the Pope released his much-talked-about encyclical.

Chris Olah: There is a real possibility that AI will displace human labor at a very large scale. If that happens, supporting those displaced will be a moral imperative of historic proportions. How will we ensure that the gains of AI are shared globally? We do not have a mechanism for this. It is an unsolved problem, and it is the kind of problem the Church has historically refused to let the world ignore.

Brian Merchant: [chuckles]

Micah Loewinger: I heard you snickering. Why does this make you laugh? This is a serious problem.

Brian Merchant: Well, to me, this is such a great example of AI executive brain, because we have historically proven policy tools that are actually pretty good at redistributing some of the wealth and power that's concentrated among a certain group of people with everybody else. They're called taxes. Just because AI companies don't really like that idea, they don't like the concept of being taxed, doesn't mean they don't exist. There are some pretty low-hanging fruit solutions here that you could talk about.

Micah Loewinger: You read Pope Leo XIV's 42,000 words on this topic. What did you think of it?

Brian Merchant: It's actually a really fascinating and rich document. I think the important things to know about it are that Pope Leo views it as a continuation of or comment on the previous Pope Leo, Leo XIII, his famous encyclical Rerum Novarum, which was written and released in 1891. It's viewed as this transformative document in Catholic history as it's really marks the beginning of the Catholic tradition of social teaching.

It's primarily concerned with the impact of the Industrial Revolution and how the industrialists are treating workers in the wake of this time of great tumultuous change, the rapid concentration of wealth, the rise of machinery, and so many human workers falling under the gears. It makes this argument that in order to truly be a good Christian, to have the time and the space and the spiritual capacity to live with God, you also have to have your material conditions met, right? You can't be suffering.

Therefore, the Church decides that every human is worthy of dignity. Every human worker should be protected, even if that means state intervention to get there. Pope Leo XIV takes a lot of this framework and then moves it into the modern era, the AI era. Let me even just quote him here. Leo writes that the fundamental dignity of each person is neither acquired nor earned, nor does it need to be justified. It's not, "Oh, is he working hard enough, or is he better than AI? Is he worse than AI? What's his productivity output?" No, the dignity is inherent. It's pretty powerful stuff.

Micah Loewinger: Timnit Gebru, who's an AI ethics expert, she previously worked at Google. She wrote on LinkedIn, calling this PR event "Vatican washing," like greenwashing, allowing Anthropic to robe itself in the holiness of the Pope's righteous criticism of this industry and the harm it might cause. She wrote, "The Vatican could have told Anthropic to stop stealing data, exploiting labor, killing the environment, deceiving us with anthropomorphic designs, and lying about product capabilities. Instead, they partnered with them, like partnering with the Sackler family to discuss the harms of Oxy."

Brian Merchant: Yes, she did not hold back there. I think that's absolutely correct. They chose to have Anthropic there and speak out. An AI executive, a billionaire, presenting the AI company's take on the whole affair. They could have had a worker who's lost his job to AI, or a content moderator who has developed PTSD, looking at some truly horrific stuff that AI is outputting, but is working for dollars a day to make these systems work.

It was a huge public relations coup for Anthropic, getting to be there, co-signing this document of great moral authority. I think it should bother us that just days later, Anthropic ran home and announced the biggest funding round yet, and that their company's now worth almost $1 trillion, on the back of all of this performative moral hand-wringing.

Micah Loewinger: Do you think there's political appetite in the United States for the type of political intervention that the encyclical calls for?

Brian Merchant: If it's not here right now, it's on the way. I think there's a rapidly growing demand for political action against this yawning inequality that we're seeing get entrenched.

Micah Loewinger: Yes, I think of the image of an American executive at a commencement speech preaching AI to classes of graduating college seniors.

Gloria Caulfield: The rise of artificial intelligence is the next Industrial Revolution.

[crowd boos]

Micah Loewinger: The boos they get.

[crowd boos]

Gloria Caulfield: What happened? Okay, I struck a chord. May I finish?

Micah Loewinger: It just feels like both a temperature check of this moment, but also foreshadowing of the political battles to come. How do you think we should be talking about our future that doesn't just do the kind of PR work that the companies are looking for?

Brian Merchant: There are a lot of great critical journalists covering this stuff, folks like 404 Media and WIRED. I have a series of work that I do called AI Killed My Job. The whole goal is to assess the claims of the AI CEOs by talking to the workers, right? That's how we know what's actually happening. Most of the time, if you talk to someone who's at a workplace that has been heavily saturated with AI, they'll have a much more nuanced story to tell than just, "Yes, it took my job." You hear stories about how, "Yes, they talk about AI a lot, and they force us all to use it, but we don't actually see a whole lot of gains."

In fact, there's been a lot of talk in just the last couple of weeks about whether or not AI is actually providing the return on investment that was promised. You have big firms like Uber saying, "I think we're going to scale back." Amazon saying, "I think we might have to scale back." We also need, I think, more focus on the financial balance sheets about what's actually going on behind the scenes here. They've gotten so much money by telling these wild stories without having a concrete business plan to put forward, to the extent that Sam Altman has said early on--

Sam Altman: We have no idea how we may one day generate revenue. We have made a soft promise to investors that once we've built this generally intelligent system, basically, we will ask it to figure out a way to generate an investment return for you."

[laughter]

Brian Merchant: That was the plan.

Micah Loewinger: [laughs]

Brian Merchant: People did laugh at him in the room, and he said, "No, no, no."

Sam Altman: You can laugh. It's all right, but it is what I actually believe is going to happen.

Brian Merchant: For all of the fascinating technology and all of the breakthroughs and all of the disruption that this has caused, it is built, first and foremost, on this story. We need journalists who are going to continue to dig into what lies beyond the story.

Micah Loewinger: Brian, thanks for doing this work.

Brian Merchant: Thanks so much for having me.

Micah Loewinger: Brian Merchant is author of the book Blood in the Machine.

Brooke Gladstone: Coming up, did AI write that?

Micah Loewinger: This is On the Media.

[music]

Brooke Gladstone: This is On the Media. I'm Brooke Gladstone.

Micah Loewinger: And I'm Micah Loewinger. Not long after the release of the Pope's encyclical, a rumor started to spread on social media that his missive on AI may have been written using AI. The fact-checking site Snopes could not find conclusive evidence to support this claim, but it did reference scan results from Pangram Labs, an AI detection tool, which found that the encyclical was 94% human, 4% AI-generated, and 2% AI-assisted.

If those results are accurate, then maybe someone on the Pope's staff fed parts of his essay into, say, Claude for editing help. In this moment, as it's becoming harder to tell what's real or fake, human or AI, more people are turning to tools like Pangram. In March, a researcher fed a New York Times Modern Love column by Kate Gilgan into the app and alleged that the piece was more than 60% AI-generated. One writer for The Atlantic decided to dig deeper.

Vauhini Vara: I eventually spoke to the author of the column over text message.

Micah Loewinger: Vauhini Vara is the author of the book Searches: Selfhood in the Digital Age. She's been tracking how AI has been creeping into reporting, writing, even literature.

Vauhini Vara: She told me that she hadn't actually copied and pasted anything from an AI model into her work, but then she went on to say that she did utilize AI as a tool for things like inspiration and guidance and correction. Then she gave me this whole list of five AI products that she'd used while writing the piece, and she called AI a collaborative editor.

Micah Loewinger: You reached out to The New York Times about the fact that they had published something that, at the very least, was written with a heavy assist from AI tools. How did they respond?

Vauhini Vara: The New York Times said that if people used generative AI in a substantial way that needed to be disclosed to readers. I asked this spokesperson, who I was in touch with, whether the level of AI use in that essay rose to that level that required disclosure. She didn't directly answer the question. Eventually, many weeks later, The Times actually did change its policies. It introduced this new AI policy for freelancers, making clear that they were expected to largely refrain from using AI.

Micah Loewinger: Here's another example of an AI reckoning, the horror novel called Shy Girl. It had garnered some buzz before it was released by Hachette Book Group, until a New York Times article made the case that, again, AI tools may have been involved.

Vauhini Vara: In that case, the author of that book did tell The Times that she had not used AI to write Shy Girl, but that she had hired an acquaintance to edit an earlier version of the book. That acquaintance had used AI, and then Hachette canceled publication.

Micah Loewinger: Okay, let's talk about another case study where, again, this AI detection tool, Pangram, was used. Last year's Commonwealth Short Story Prize.

Vauhini Vara: People again on social media started noticing that this one story that was published online on the website of the magazine Granta, which is very prestigious, seemed to just reek of AI use. Again, people ran the story through Pangram. In that case, Pangram believed that the story was 100% AI-generated, but that was just the beginning of it because that story had actually been a regional winner of this prize, called the Commonwealth Short Story Prize. There were four other regional winners. People ran those stories through Pangram. One other story was flagged as 100% AI-generated. A third story was flagged as including significant amounts of AI-generated material.

Micah Loewinger: What made this instance different from the other two we talked about is that these writers said that they didn't use AI. I want to ask you more about these AI detection tools like Pangram. How does Pangram work? How accurate is it? Do we know?

Vauhini Vara: It essentially shows lots of different examples of human writing and AI-generated text to an AI model. That model learns to distinguish the difference between what human writing looks like and what AI-generated writing looks like. It isn't a matter of this technology, like counting em dashes per paragraph, or tallying the number of really bad metaphors.

Micah Loewinger: Which, to a human eye, might be a sign of AI writing.

Vauhini Vara: Exactly, but Pangram, in particular, as well as another product called GPTZero, have been found by independent research from the University of Chicago to actually be very accurate. There's almost a 0% chance that something's going to be flagged as AI if it's really human-written. The product does have higher false negative rates, which can be between 2% and 4%, according to this University of Chicago research, which means that it's somewhat more likely that it's going to take something that was actually AI-generated and flag it as human-written.

Micah Loewinger: Interesting. In some of your own reporting, you have relied on Pangram's analysis and, more specifically, the work of a Maryland PhD student and researcher named Jenna Russell. She and other researchers from Pangram released a preprint, meaning you can read it online, though it hasn't been peer reviewed yet, which scanned 185,000 articles from 1,500 newspapers.

The team claims that 9% of those newspaper articles were either AI-generated or mixed, and op-eds specifically were six times more likely to have AI's fingerprints. In an op-ed published in The Wall Street Journal in April, an editor at the paper, James Taranto, dug into that Pangram research paper, which he says erroneously flagged three op-eds in The Wall Street Journal as AI-generated.

That term matters because two of the accused writers said they used AI to revise some of their work, but he wrote, "It's inaccurate and unfair to characterize work produced in the manner described by any of these authors as AI-generated. The Pangram tool has no ability to discern a writer's work process. It merely looks for linguistic patterns that tend to correlate with AI output."

Vauhini Vara: I wouldn't come to the same conclusion as the writer of that piece, but it is true that to flag something as AI-generated does not reflect the nuance of how people actually use AI products, right? When you see that something is flagged as 100% AI, for example, I think what we picture is that somebody went to ChatGPT and said, "Write me an 800-word Modern Love column about a bad breakup."

It may be the case, actually, that people are giving the AI models very rough drafts or outlines of the material, and then relying very heavily on the AI chatbot to suggest language or refine their language. An author might feel that it was actually their work because it was their ideas, or an original draft came from them. I think that's a fair conversation to have, but I think it raises this other problem, which is that we don't really have clear standards about what we mean when we talk about AI reliance.

Micah Loewinger: In all of the examples that we've discussed, shaming seems to play a role. I guess I'd love to hear what you would say to people who don't necessarily believe that all of these writers did something wrong.

Vauhini Vara: Listen, I think it's a matter of disclosure and transparency. We can imagine a world in which The New York Times or The Atlantic or New York Public Radio have very clear policies about AI that they post on their websites, for example. Then, in every article or opinion piece, there's a disclosure explaining exactly how AI was used in producing that piece. That way, people can know, and they can make an informed decision about whether they want to read the piece at all, how much they trust it.

Micah Loewinger: There was a really interesting piece in Poynter last month by a book critic named Rachel Meghan. She had written about Shy Girl before its release and had given it a rave review, before reading that it was 78% AI-generated. She wrote that the experience was embarrassing. It made her question her judgment as a critic. It made me wonder. If she, a fan of horror, found such value in this book, why then does the realization that AI played a role undo all of that?

Vauhini Vara: For me, it has to do with what literature is for. When we view it through that lens, if I, as a reader, appreciated it, isn't it maybe okay? It's a consumerist lens to use to think about literature, I think. We are customers, and all we need is to be given products that we appreciate, right? There's another lens through which to look at it, which is a political lens, which considers literature, but also journalism, and also just communication in general, to be this crucially important political act. There's something politically powerful about using language to serve our individual and collective purposes as people in the world. Something really dangerous, I think, about accepting language that's provided to us by big technology companies for which we have to pay in some way or other, right?

Micah Loewinger: We're electing to give up our agency as thinkers and individuals when we let them talk for us.

Vauhini Vara: Exactly. The thing that makes us human as a species is our ability to use this really specific form of language that only we have in order to connect with one another. Yes, it's not really about like, "Should we do it, or should a robot do it?" It's about, "Should we do it, or should we relinquish it to products built by really rich, really powerful technology companies that are built to further enrich and further empower those companies?"

Micah Loewinger: Vauhini, thanks for coming on the show.

Vauhini Vara: Thank you for having me.

Micah Loewinger: Vauhini Vara is a contributor for The Atlantic. She's the author of the recent piece, This Literary AI Scandal Changes Everything.

Brooke Gladstone: Coming up, an in-real-life tour through the Epstein files.

Micah Loewinger: This is On the Media.

[music]

Micah Loewinger: This is On the Media. I'm Micah Loewinger.

Brooke Gladstone: And I'm Brooke Gladstone. A few months ago, the Justice Department released on its website its last tranche of Epstein files, a tangled thicket of PDFs. Some 30 public figures, mostly foreign politicians, businessmen, academics, and lawyers, have faced some sort of reckoning for their involvement with Jeffrey Epstein as a result of the files, but America's most powerful seem to have been left relatively unscathed.

Representative Madeleine Dean: The President has lied about being on Epstein's plane, and the unredacted files prove that. There's a lot in here.

Brooke Gladstone: Pennsylvania Rep. Madeleine Dean questioning Acting Attorney General Todd Blanche earlier this week as part of the House of Representatives' investigation into the federal government's handling of the Epstein files. It's worth noting here that while some three-and-a-half-million documents in the government's files on Epstein have been made public, roughly two and a half million have not.

Representative Madeleine Dean: There's also this set of files. This is investigation into the potential co-conspirators of Jeffrey Epstein. It's all covered up. The American people are not stupid. They know that when members of Congress have to go in and actually unredact, try to find the truth for these victims, something is corrupt. Something is corrosive.

Brooke Gladstone: On a sticky day last month, I stepped into the cool interior of what had been an art gallery to behold the files in real life, at least those we've been authorized to see.

David Garrett: In late March, we downloaded all of the Epstein files from the Department of Justice website, printed them out, bound them into volumes, and brought them here.

Brooke Gladstone: David Garrett is a luxury wine entrepreneur from Michigan and the co-founder of the nonprofit Institute for Primary Facts. He organized this exhibit to shine a spotlight on the decades-long relationship between Jeffrey Epstein and Donald Trump, its synergies, and synchronicities. The exhibit was open to the public in New York City for several weeks in May.

David Garrett: 3.5 million files ends up being about 3,500 volumes of 800 pages each.

Brooke Gladstone: They're all in this room?

David Garrett: We couldn't fit them all in here, so we have a few downstairs, but yes, 17,000 pounds of evidence.

Brooke Gladstone: At that moment, I was standing in front of a bookcase filled with volumes and stark white binders. I pulled out volume 2,742 as released by the Department of Justice of the United States, The Donald J. Trump and Jeffrey Epstein Memorial Reading Room, published by the Trumpsonian. [chuckles] I laughed.

David Garrett: This all started when the Smithsonian took down the exhibits of Trump's impeachments. I lost it a little bit. I'd been listening to too many podcasts and yelling at the TV for 10 years. When the Smithsonian decided to bow to the Trump administration and rewrite history, I thought, "We need someplace that's going to actually preserve the truth."

Brooke Gladstone: I think a lot of people might think that, wow, it really takes a long time to Xerox all those files. We know that it took a long time for Ellsberg to get the Pentagon Papers Xerox'd, but that was a walk in the park compared to this.

David Garrett: There were really two problems getting everything printed. The first one was technical. All of these files are heavily redacted. When you redact a digital file, they go over the redacted material three times with full black. It's completely gone. The problem is that so much ink that if you put it through a printer, it'll tear the paper. We had to go through all 3.5 million files and correct each pixel that was 300% black to a format that could be printed.

The bigger problem was really political. The first printer that we sent this project to came back and said, "We work with a lot of lawyers. We're really worried that if they find out that we printed this, they're not going to work with us anymore." A second printer, one of their biggest investors, was a Trump supporter. We finally found some printers that are super courageous, and we got very lucky.

Brooke Gladstone: The locations?

David Garrett: We had a very bad day when we lost both the printer and our first location on the same day. We literally sat down and said, "Is this too hard?" We probably looked at five different places.

Brooke Gladstone: They turned you down?

David Garrett: Yes, a bunch of them turned us down. Yes. This exhibit, the reason that Donald Trump's name is on it is really for two reasons. One is what he shares with Jeffrey Epstein is dozens and dozens of allegations of sexual assault. They're both convicted felons, so putting their names together really makes sense.

Brooke Gladstone: Trump's name is cited-

David Garrett: -more than 38,000 times throughout the documents. People can draw their own conclusions from that, but my point is this. He shouldn't be influencing the investigation into these crimes. That is a corrupt act, him slowing that investigation. That's the purpose of having his name on this.

Brooke Gladstone: Now, I assume you've looked at a lot of these documents, right?

David Garrett: Too many.

Brooke Gladstone: Will you show me one that struck you?

David Garrett: Oh, I-- Well, here's what I'll tell you. One time last week, I was giving an interview. I just pulled a random book off the shelf and started leafing through it to show them that it was real. It was page after page of little girls with no clothes on. I had to put the book back, and I had to sit down, and so I don't do that anymore. I was here with one of the survivors a couple of days before we opened. We helped her to find the books where she's mentioned. We pulled them down, and I sat down with her downstairs. She showed me her name, her address, her birthday, her Social Security number.

Brooke Gladstone: Things that were supposed to be redacted?

David Garrett: A picture of her driver's license. She was 17 when she did the FBI interview. She talked about some of her friends that she had told about it. Those friends' names are unredacted. It is criminal what the Department of Justice has done in causing additional harm to the victims and survivors and exposing them. I don't know who to see about it when the Department of Justice breaks the law.

Brooke Gladstone: He walks us towards the back of the room to display the famous picture of Trump and Epstein. Trump's got his hand on his shoulder. They're both smiling, and then there's a pretty extensive timeline. Three actually. One on top of the other on top of the other.

David Garrett: The top one is Jeffrey Epstein, his dozens of allegations of sexual assault, his felony conviction. The bottom timeline is Donald Trump and his dozens of allegations of sexual assault and his felony convictions. In the middle, you see all these famous photos of them together. We built this timeline to create context, so you would know when these photos were taken and all of the allegations of sexual assault by both men. The notes that you see that are tacked to the timeline are all from survivors that have been here over the course of the last few days. They see themselves on the timeline, and they asked if they could include themselves.

Brooke Gladstone: This one here from Marina Lacerda (Minor Victim 1), "My first time in Epstein's house, I was only 14 years old. He gained my confidence and groomed me till he was able to rape, abuse, and take my childhood away. Today, I am 37 and reclaiming my power, and I encourage every woman not to forget that power is always within. Love, Marina." This happened to her in 2002, alongside a profile of Epstein in New York Magazine, Trump said, "I've known Jeff for 15 years. Terrific guy, a lot of fun to be with. It's even said that he likes beautiful women as much as I do, and many of them are on the younger side." Same year as Marina's encounter.

David Garrett: When she was 14.

Brooke Gladstone: Yes.

David Garrett: This is in 2005, the same year the federal probe into Epstein started. Dani writes here that she was trafficked to Epstein in 2004 at 17 years old. "Jeffrey used dance, my ballet background, and my mom's brain tumor against me to get what he wanted. Demand from the powerful still exists. We have to make it safer."

Brooke Gladstone: This is around the same time that Trump wrote that birthday card?

David Garrett: Yes, she put it on the timeline right above Trump's famous birthday card.

Brooke Gladstone: 2000, the same year, where there's the famous foursome photo of Trump, Melania, Epstein, and Ghislaine Maxwell photographed together at Mar-a-Lago. This is when Virginia Giuffre was recruited at Mar-a-Lago. She was 16 years old. She became one of the most prominent voices against Epstein. Later, she killed herself. Lisa Phillips, around the same time, brought by boat from Tortola. She writes in this note on your wall, she met Prince Andrew "on the same day Epstein," she writes, "abused me."

David Garrett: Yes, so everything that's happening now with transparency and accountability, as flawed as it is, is because of these amazingly brave women who stood up and refused to go away. The pictures that you see here are them speaking out in front of Congress to get the Epstein Files Transparency Act passed, and then two of these photos are from the very famous public service advertisement that ran during the Super Bowl, with all of the survivors holding up pictures of themselves the year they met Jeffrey Epstein.

Survivor 1: After years of being kept apart, we're standing together.

Survivor 2: Standing.

Survivor 3: Standing.

Survivor 4: Together.

Survivor 1: Because this girl deserves the truth.

Survivor 5: Because she deserves the truth.

Survivor 6: Because we all-

Survivor 7: -deserve the truth.

Brooke Gladstone: On the timeline, there's a girl. She looks maybe 13 or 14, smiling for her class photo. This was the year she met Epstein, 1991. She leaves a note, "Jess Michaels, raped by Jeffrey Epstein, brought to him by a woman who would go on to work for Epstein for 18-plus years."

David Garrett: What we have in the middle is a--

Brooke Gladstone: Next, David Garrett brings us to a draped box. Inside, there are 1400 electric candles.

David Garrett: Each one of those candles represents one of the victims of Jeffrey Epstein. Each one of them is a life.

Brooke Gladstone: You found that people looking at it had a tendency to start counting.

David Garrett: Yes, I think they start counting first to just say how many are there. Some people will stand here for 20 minutes or a half an hour. What we're surrounded by are 3.5 million files. It's the problem with big numbers. How do you know what 3.5 million files looks like? Well, now, you can see it, but the same thing is true with 1,400 victims. How do you know what 1,400 little girls looks like? 10 is a volleyball team. 20 is a lacrosse team. 1,400 fills a theater.

Brooke Gladstone: Downstairs, the walls are blanketed with so many more handwritten notes, but these are written by the guests.

David Garrett: This was our overflow room originally, where we have some more bookshelves and a few hundred last books. We created a space down here where people that come through, if they want to call the Department of Justice, they can sit at that table and call and ask, demand that all of the files be released, and that they be properly redacted. Then, here at this table, people leave notes. We had one corkboard. We thought we'd only need one for the whole time that we were here. There are now maybe 20 corkboards.

Brooke Gladstone: Here's one that says, "We live in a prison with free monsters. We are not things. All this happened in the 'best country in the world.' Are we great yet?"

David Garrett: A lot of people that come through are survivors of sexual abuse or sexual assault themselves. A lot of women need a moment after seeing the exhibit upstairs and want to express themselves down here. We've seen dozens and dozens of survivors come through this space.

Brooke Gladstone: There's one that says, "Why do men in power keep trying to protect us from imaginary threats and ignoring the real ones? Burn it all down." You call this enterprise a pressure campaign. How so?

David Garrett: A year ago, when the authors of the Epstein Files Transparency Act were contemplating that piece of legislation, everybody thought, "It'll never pass. It'll never be law," but because these incredibly brave women went to Capitol Hill and demanded that this law be passed, because millions and millions of people called their congresspeople, it passed the House with only one no vote.

It passed the Senate unanimously, and the President signed it. What that means is that if we work together, democracy can still work. That's what I hope we can do. A small part of here get people excited. We've already done maybe 1,500 calls to the Department of Justice. We hope to demand that all the files be released, that they be properly redacted this time, and that there is real transparency that drives towards real accountability.

Brooke Gladstone: The Institute for Primary Facts is primarily an anti-corruption organization. Is that wrong?

David Garrett: It's a pro-democracy, right? I think that's the same thing. Corruption is the tool of autocrats to kill democracies.

Brooke Gladstone: Primary facts seem to have much purchase in the current political climate. People seem to be immune.

David Garrett: I'm not sure if it's the people. I think it might be the medium. You make more money, the more attention you get on social media. The things that get the most attention sometimes aren't true. You just have no idea when you're looking through your phone. When I say the Institute for Primary Facts, what I'm hoping that we can do is use, in real life, use tactile experiences. Get away from your phone and come in and see what we've done here. It's really hard to ignore. You can't scroll past it when you're walking through this room.

Brooke Gladstone: So many people are able to, "Tsk, tsk," at what happened with Jeffrey Epstein and move on. You clearly can't. Just talking about it made you something that the listeners couldn't see, but really upset. Why can't you pass it by? I have two daughters. How old are yours?

David Garrett: I have a daughter who's 26, and one who's 15. I imagine that I'll have granddaughters someday. There have always been monsters. There have always been monsters. For 250 years, we have the rule of law. The rule of law meant that we could feel confident that, regardless of who the perpetrator was, we could hold them accountable. If we could hold them accountable, then there would be a deterrent for people to do it in the future. Well, what is the deterrent now? They've all gotten away with it. I don't think that's providing a safe environment for my daughters and granddaughters. Let's have some real accountability.

Brooke Gladstone: Look, I've been a reporter for a long time. I've often told young reporters that it's like water dropping on a stone to change things. You need a billion drops to even slightly redirect the flow. Are your hopes for this reasonable?

David Garrett: Maybe. Even if they're unreasonable, real change is a small group of people trying to do impossible things. That's the only way that change has ever happened. I think that a lot of people are hungry for change, so let's do it.

Brooke Gladstone: Thank you so much.

David Garrett: Thank you.

Brooke Gladstone: While Garrett was guiding us through the tour, a woman in a long navy dress slipped into the exhibit. He introduced her as Andrea Sterling, an online content creator and also a survivor of one of Epstein's notorious assaults on minors. She agreed to sit down with me downstairs amid the spillover volumes and all those encouraging corkboards and tell me what happened to her.

Andrea Sterling: I was 17, and I recently moved from Colombia. A friend of mine casually invited me. He was like, "Hey, I have this gig. There's this rich guy, and I give him massage." There were no details. It was easy for me to accept, too, because she always invited me to very normal events. Barbecues, friends that were our age. Very normal.

When she invited me, it was a little strange. To be honest with you, I trusted it. He clearly had a technique. Two cleaning ladies opened the door for me and his residence here in New York City. They put me in an office. In the office, you can acknowledge all the pictures. Now that I'm older, I'm like, "Oh, they did it on purpose for you to feel safe," like, "Oh, he's with politicians. What bad can happen?"

Brooke Gladstone: Which is the trap.

Andrea Sterling: That's a trap.

Brooke Gladstone: Do you think famous people won't behave that way?

Andrea Sterling: I thought so. I go with the two girls who brought me to the massage room. Quickly, I realized that it's not all about a massage. It was not a massage. He just continued, and you're trapped. Some girls freeze. I was fighting on it, but then froze. It was a disgrace.

Brooke Gladstone: You felt shame. You felt shame?

Andrea Sterling: Unfortunately, it's hard to explain. I don't understand. How did I carry that? That didn't belong to me and didn't belong to any of the girls, so I leave whenever I could leave. I stopped talking to the friends that brought me there. Now, I understand that they were the victims, too. They were minors. They were even younger than me.

Brooke Gladstone: Then, there's this exhibit of the Epstein files. How did you get here? How did you happen to walk in? You never even made your name public.

Andrea Sterling: No. In 2018 or '19, when I saw him, I recognized him. I didn't want to get involved. I wasn't sure. I saw a lot of the survivors already talking. I thought, "These girls are so brave. They're already doing it." Something's going to happen, right? Keep on waiting and waiting and waiting, and then I'm like, "Wait, this is getting darker. I have to step in." I create content in the cannabis industry, in jewelry, and traveling.

I create a lot of fun and comedy content. I felt like a coward. Now, I think it's time to step in and be part of the noise. That's what I'm doing right now. Then, they contact us about this amazing project. I think this is very helpful because it helps you understand the magnitude. These are only half of the files. These are the vanilla files, even though they're very dark, but I think they gave us the safer files. Imagine the other ones.

Brooke Gladstone: I understand that you allowed your name to be used here for the first time.

Andrea Sterling: I opened up in Latin America only. Telemundo, Univision. In the United States, only from here, yes, today in public.

Brooke Gladstone: Wow, and so your strongest reaction to this enterprise was-

Andrea Sterling: -putting my name in the timeline, and to see how many people are with us. They take their time to actually come here and write a note to encourage us to keep going. A lot of survivors, they've been talking since 1996, 1998.

Brooke Gladstone: The shame you mentioned?

Andrea Sterling: The shame is gone. That's why I decided to talk to shame. You understand when you're an adult.

Brooke Gladstone: You've been an adult for a while, but it didn't--

Andrea Sterling: Yes, it didn't hit me until recently. It has been a process. I did have some therapy prior, worked on some of the things, but what am I going to be ashamed of? I wasn't the 55-year-old hiring minors to hire older minors, so the shame is not mine, yes.

Brooke Gladstone: We're very lucky you walked in.

Andrea Sterling: Oh, no, thank you. I'm very lucky to meet you. [chuckles]

Brooke Gladstone: The Donald J. Trump and Jeffrey Epstein Memorial Reading Room has closed in New York, but it's next headed to Washington, DC.

[music]

Micah Loewinger: That's it for this week's show. On the Media is produced by Molly Rosen, Rebecca Clark-Callender, and Candice Wang. Travis Mannon is our video producer.

Brooke Gladstone: Our technical director is Jennifer Munson, with engineering help from Jared Paul. Eloise Blondiau is our senior producer, and our executive producer is Katya Rogers. On the Media is produced by WNYC. I'm Brooke Gladstone.

Micah Loewinger: And I'm Micah Loewinger.

 

Copyright © 2026 New York Public Radio. All rights reserved. Visit our website terms of use at www.wnyc.org for further information.

New York Public Radio transcripts are created on a rush deadline, often by contractors. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of New York Public Radio’s programming is the audio record.