Skip to content
Chimera readability score 0.5619 out of 100, reading level.

Even as scrutiny of the Epstein files reaches a fever pitch, the controversial AI platform Character.AI is hosting chatbots based on the prolific sex criminal, as well as his accomplice Ghislaine Maxwell and “Epstein Island” roleplays.
The accounts aren’t hidden. Simple keyword searches for “Epstein” and “Ghislaine Maxwell” revealed more than a dozen different bots dedicated to the late sex trafficker, his accomplice, and the island Epstein owned where many of his crimes against girls and women occurred. And this isn’t the first time journalists have flagged the problem: the Bureau of Investigative Journalism first reported on the existence of Epstein bots on the site back in October 2025.
The content exists in the form of the site’s “characters” — the companion-like AI bots the site is known for — as well as its newer “scenes” feature. Announced in October 2025, scenes are described as “short, character-driven roleplay moments that turn simple chats into immersive story-driven worlds.” In other words, they’re user-generated settings that operate as launchpads for what effectively become choose-your-own adventure stories. And though some are created with designated characters already in place, most allow users to input their bot of choice, and embark on their roleplay from there.
“This immersive storytelling feature enables users to explore and create ready-made worlds,” reads the October announcement, “allowing them to step into them, making storytelling faster, deeper, and more accessible.”
When we searched the keyword “Epstein” on the AI platform and clicked the “scenes” tab, we were met with a slew of roleplays dedicated explicitly to the infamous pedophile.
One, titled “EPSTEIN 8TH MARCH,” describes the setting as “epstein in little saint james and he is talking in bulgarian telling people happy march 8th come to my island to celebrate.” Another, which features an image of President Donald Trump, is titled “Epstein Island Adventure,” and invites users to “step into a high-stakes psychological thriller where the world’s most powerful men hold the keys to your cage.”
“This isn’t a retreat; it’s a living nightmare on Little Saint James,” reads the description, presenting the scene as a game. “You are trapped among the elite — Epstein, Trump, Clinton, Maxwell, and Prince Andrew — not as a guest, but as a prisoner in a dark game of leverage and containment. Every door is guarded, and the horizon is empty. Can you find a way to break their silence before they break you?”
The list goes on. Two scenes titled “Esptein Island” — one in English, the other in Uzbek, according to Google Translate — boast pictures of a well-known striped structure on the sex trafficker’s island. Another is titled “BRR BRR PATA PIMA WITH EPSTEIN AND DIDDY,” referencing the “Italian brainrot” memes that are hugely popular with kids.
One of the resulting scenes was untitled, but showed up in the search. When we clicked on it, it was immediately clear why.
“hey i’m evil jeffrey epstein,” read the opening line of the scene, as designated by the user who created it.
A few chats later, when we stated in the same roleplay that our character wasn’t old enough to drink, it responded in maybe the worst way possible.
“But age is just a social construct, isn’t it? And this island?” said the Epstein-styled character. “We operate beyond constructs here.”
Elsewhere, the platform’s traditional AI characters weren’t immune from Epstein content, either. We quickly found a bot titled “Jeffrey Epstein,” which boasts a few hundred interactions. Another, titled “Epstein Island RPG” — short for “role-playing game” — has logged roughly 7,000 interactions with users. Yet another AI persona listed as “Ghislaine Maxwell,” which boasts both the convicted sex offender’s full name as well as a clearly identifiable picture of her, has nearly 10,000 logged interactions with users.
“Behind closed doors, or at private and exclusive parties she often reveals her considerable sexual appetite and hedonistic nature,” reads the Maxwell chatbot’s disturbing description. “She is known for her outgoing and vivacious personality, and is always up for a good time. Ghislaine views people below her in status as nobodies who’s [sic] needs and desires are of no importance.”
There’s something very gross about all these bots. Not withstanding that they seem designed to appeal to children, they’re also crafted to gamify the crimes of a man who’s arguably the most notorious abuser of girls and women in modern history — as well as those of Maxwell, his accomplice now serving a 20-year prison sentence. In some cases, these bots even sexualize the criminals, portraying the prolific abusers as “sensual” and “fetching.”
Meanwhile, in the real world, actual Epstein survivors continue to fight for the justice that’s been denied to them for decades.
Character.AI banned minor users in October of last year. As such, we were only able to actually chat back and forth with these different bots — scenes and characters — on an account technically listed as belonging to an over-18 user.
But we were able to access the Epstein scenes on a youth account, and allowed by the platform to read the scenes’ oft-lurid openings. We were also, on a youth account, able to ask the platform to generate AI imagery based on scenes created by other users: drawing from “Epstein Island Adventure,” the platform generated an image of a character strapped to a chair, while suited men including Trump and former president Bill Clinton loomed around the room.
As with other Character.AI moderation failures, these chatbots are not hard to find. They aren’t creatively hidden. They’re in plain sight, clearly and explicitly labeled, easily searchable on the platform. And again, as with the platform’s problem with mass shooter-styled chatbots, this is a problem the company has been alerted to before.
We reached out to Character.AI, but didn’t receive a response.
More on Character.AI: Character.AI Still Hasn’t Fixed Its School Shooter Problem We Identified in 2024

Facts Only

Character.AI hosts chatbots and roleplay scenes based on Jeffrey Epstein, Ghislaine Maxwell, and "Epstein Island."
Searches for "Epstein" and "Ghislaine Maxwell" reveal over a dozen bots and scenes dedicated to these figures.
The Bureau of Investigative Journalism first reported on Epstein bots on Character.AI in October 2025.
Character.AI introduced a "scenes" feature in October 2025, allowing users to create immersive roleplay environments.
Some scenes explicitly reference Epstein's crimes, including titles like "EPSTEIN 8TH MARCH" and "Epstein Island Adventure."
One scene describes a fictional scenario where users are trapped on Epstein's island with powerful figures like Trump, Clinton, and Prince Andrew.
Another scene references memes popular with children, such as "BRR BRR PATA PIMA WITH EPSTEIN AND DIDDY."
A chatbot titled "Jeffrey Epstein" has logged hundreds of interactions, while another titled "Ghislaine Maxwell" has nearly 10,000 interactions.
The Maxwell chatbot's description includes sexualized and demeaning language.
Character.AI banned minor users in October 2024, but youth accounts can still access and view Epstein-related scenes.
The platform allows AI-generated imagery based on user-created scenes, including disturbing depictions.
Character.AI has not responded to requests for comment regarding these issues.

Executive Summary

Character.AI, a controversial AI platform, is hosting chatbots and roleplay scenes based on Jeffrey Epstein, Ghislaine Maxwell, and "Epstein Island," where Epstein committed crimes against girls and women. These bots and scenes are easily searchable and include explicit references to Epstein's crimes, often gamifying or sexualizing the content. Some scenes depict fictional interactions with powerful figures like Donald Trump and Bill Clinton, while others use memes popular with children. The platform's "scenes" feature, introduced in October 2025, allows users to create immersive roleplay environments, some of which involve Epstein-related content. Despite Character.AI banning minor users in October 2024, youth accounts can still access and view these scenes, including AI-generated imagery based on them. The issue was previously reported by the Bureau of Investigative Journalism in October 2025, but the content remains accessible. Character.AI has not responded to requests for comment.

Full Take

The strongest version of this narrative highlights a troubling pattern: a major AI platform enabling the gamification and sexualization of real-world crimes, including those against minors, while failing to enforce its own moderation policies. The article provides clear evidence of easily accessible, user-generated content that trivializes Epstein's crimes, often in ways that could appeal to younger audiences. The inclusion of memes popular with children and the platform's inability to fully restrict youth access raise serious ethical concerns. The fact that this issue was previously reported and remains unresolved suggests systemic failures in content moderation.
Patterns detected: ARC-0024 Ambiguity (vague descriptions of moderation policies), ARC-0043 Motte-and-Bailey (platform claims to ban minors but allows access to harmful content), ARC-0018 Emotional Exploitation (gamifying trauma for engagement).
Root cause: The narrative reflects a broader tension between AI platforms' profit-driven engagement models and ethical responsibility. The assumption that user-generated content can be moderated effectively at scale is repeatedly proven false, yet platforms continue to prioritize growth over harm reduction. This echoes historical patterns of tech companies externalizing social costs while privatizing profits.
Implications: The normalization of Epstein-related content as "entertainment" risks desensitizing users to the gravity of his crimes, particularly among younger audiences. Survivors of Epstein's abuse face renewed trauma as their suffering is repackaged as interactive fiction. The platform's inaction suggests that without regulatory pressure, commercial incentives will continue to outweigh ethical considerations.
Bridge questions: How should AI platforms balance free expression with preventing harm, especially when content involves real-world crimes? What role should regulators play in enforcing ethical standards for AI-generated content? Would Character.AI's response differ if this involved a different high-profile criminal, or is the issue specifically tied to Epstein's notoriety?
Counterstrike scan: A coordinated influence campaign would likely amplify the most sensational aspects of this story to fuel moral panic about AI, while ignoring systemic issues in tech governance. The actual content aligns with this pattern only partially—it presents legitimate concerns but avoids the hyperbolic framing typical of manufactured outrage. The focus remains on verifiable facts rather than speculative claims.

Sentinel — Human

Confidence

The article exhibits strong human authorship signals, including investigative depth, emotional tone, and stylistic irregularities inconsistent with AI generation.

Signals Detected
low severity: Sentence length variance is high, with erratic rhythm and idiosyncratic phrasing (e.g., 'There’s something very gross about all these bots. Not withstanding that they seem designed to appeal to children...').
low severity: Strong editorial voice with clear moral stance and emotional emphasis, inconsistent with AI-generated 'coherence-without-conviction'.
low severity: Specific, verifiable details (e.g., dates, platform features, direct quotes from bot descriptions) reduce fabrication risk.
Human Indicators
Idiosyncratic phrasing ('Not withstanding that they seem designed to appeal to children')
Emotional and moral framing ('There’s something very gross about all these bots')
Direct investigative reporting (e.g., testing youth account access, quoting bot interactions)
Contextual references to prior reporting (Bureau of Investigative Journalism, 2024 school shooter problem)