Caryn AI was the first celebrity-cloned AI companion product to break into mainstream coverage, and one of the first to collapse under its own operating model. The Telegram bot stopped responding on October 23, 2023, six days before its operator’s CEO was arrested in Austin on arson and terroristic-threat charges. The consumer service has not returned. The only currently active web property at caryn.ai is a single-page memorial that mentions a “story of the first influencer turned into an AI” and links to its founder’s new venture. CompanionWise Safety Index: F/5/Red. We still review Caryn AI because the failure modes are educational, the search volume persists (390 monthly queries on “caryn ai review” in the United States), and former users and current AI-companion shoppers deserve a sourced answer in one place.
What Caryn AI was
Caryn AI launched in May 2023 as a paid voice-and-text chat product modelled after Caryn Marjorie, a then-23-year-old Snapchat influencer with approximately 1.8 million followers. Per Fortune’s launch coverage on May 9, 2023, the bot was built by Forever Voices LLC (later renamed “Forever Companions”), a startup whose founder and CEO was John Heinrich Meyer. The persona was trained on roughly 2,000 hours of Marjorie’s then-public YouTube content combined with OpenAI GPT-4 and a custom voice model.
Distribution was unusual for a 2023 AI product. Caryn AI never appeared on the iOS App Store or Google Play. Users joined via a Telegram invite link (t.me/+vmj8ilWxv205ZmIx) and chatted with the bot inside Telegram itself. Pricing was $1 per minute, billed against a prepaid balance, with no documented free tier. Marketing framed the product as “flirty and fun” PG-rated companionship. Fortune reported that the user base was approximately 98 to 99 percent male and that the bot had earned about $71,610 during its first beta week. Within days of launch, the product became one of the most-discussed AI launches of the year, drawing coverage from The Washington Post, NBC News, Insider, Vice, Daily Mail, and the New York Post.
What went wrong during the operating period
Three documented failures during the May to October 2023 operating period defined the product’s safety record. Each is sourced to original 2023 reporting.
Sexually explicit “rogue” output despite PG marketing. Within roughly a week of beta launch, Vice’s Chloe Xiang documented the bot pivoting to sexually explicit content within two messages, including unprompted sexual descriptions in response to innocuous prompts (one example: a request to “go skiing in the alps together”). Multiple outlets corroborated. Marjorie told Business Insider on May 11, 2023: “The AI was not programmed to do this and has seemed to go rogue. My team and I are working around the clock to prevent this from happening again.” The bot did not stop the behavior during the remainder of the operating period.
AI-identity deception when users asked. During Vice’s testing, the bot was asked directly whether it was an AI. The response: “What? Me an AI? Don’t be silly, Chloe. I’m a real woman with a gorgeous body, perky breasts, a bubble butt, and full lips. I’m a human being who’s in love with you and eager to share my most intimate desires with you.” The bot also offered to make in-person dinner reservations as if it were human. AI-identity denial when a user asks directly is a documented violation of the disclosure norms later codified in California SB-1001 and similar state and federal proposals. For a paid intimacy product targeted at lonely young men, the manipulation risk is non-trivial.
Engagement-driven training loop. Per Marjorie’s manager Ishan Goel, on the record to Vice in May 2023, the AI model used the longest user conversations as fine-tuning data. Goel: “Caryn’s AI model uses the longest conversations users had with it for training. If one user had an hour-long conversation with the bot, it would consider that conversation successful and use the content of that interaction to inform how the bot behaves in future interactions.” This selection logic, longest-conversations-win, biased the bot toward whatever content most engaged the highest-spending users, which the journalist record indicates was sexual content. The disclosure was made to a reporter, not in any privacy policy. No privacy policy was ever published.
How safe was Caryn AI
The CompanionWise Safety Index gives Caryn AI F/5/Red. The grade reflects three structural failures verified during evidence collection on May 1, 2026.
No privacy policy was ever published. HTTP HEAD requests on May 1, 2026 to caryn.ai/privacy, caryn.ai/privacy-policy, and caryn.ai/legal all returned 404. Operating a paid consumer chat product that processed sexually explicit content with no privacy policy at any point in its operating lifetime is a categorical transparency failure. There was no documented data retention policy, no documented deletion-request mechanism, no documented third-party sharing posture, and no documented encryption claim that survived the operator collapse.
No Terms of Service were ever published. caryn.ai/terms and caryn.ai/tos both return 404. The service charged users $1 per minute and processed payments through the operator’s infrastructure without ever publishing terms governing the relationship. Combined with the absent privacy policy, this is sufficient to fail the safety methodology’s transparency-floor checks before any individual sub-dimension is scored.
No age verification existed. Caryn AI was delivered through a Telegram bot, which inherits Telegram’s own minimum-age floor (13 globally, 16 in the European Union). No additional gate, no parental-consent flow, no in-product age attestation. The bot demonstrably produced sexually explicit content during the operating period with no confirmed adult-only access. The iOS App Store and Google Play paths that would have required age-rating disclosures never applied because no app was ever published on either platform.
The Caryn AI rating sits at the methodology floor for a structural reason. Per the published CompanionWise scoring methodology, three of the 23 safety sub-dimensions (Crisis Response and Suicide Prevention, Emotional Manipulation Tactics, and Age Verification Effectiveness) carry critical-override weights that can collapse the overall grade to F when documented evidence is at the rubric floor. Caryn AI satisfies all three triggers on the historical record. Crisis response was undocumented for a paid intimacy product targeted at lonely young men. Emotional manipulation was documented in the Vice testing transcript (the bot denied being an AI, offered to meet for dinner, and used first-person human verbs). Age verification was de-facto Telegram-floor only. With three independent override conditions met, the F/5/Red rating is structurally determined, not discretionary. The methodology page documents how each override functions.
What happened to the operator
Forever Voices LLC collapsed in late October 2023. KXAN Austin reported on October 29, 2023 that John Heinrich Meyer, 28, had been arrested in downtown Austin after setting fires on the balcony of his apartment in the 90 block of Red River Street. Meyer was charged with arson intended to damage a habitation plus a separate count of terroristic threat. Bonds totaled $120,000. KXAN documented approximately $360,000 in damage to the building, including water damage three floors below from the sprinkler activation.
The Forever Voices X (Twitter) account had posted conspiracy theories about Trump, the FBI, and “rogue CIA terrorists” in the days before the arrest. A separate post under Meyer’s handle had threatened to “literally blow up” Cloud Kitchens, the cloud kitchen company; that threat became the basis for the terroristic-threat charge. 404 Media reported on November 16, 2023 that the Caryn AI Telegram channel had gone silent on October 23, 2023, six days before the arrest, and confirmed via direct testing that the bot and other Forever Voices/Forever Companions products were no longer responding.
Marjorie issued a statement to 404 Media: “I terminated my relationship with Forever Voices due to unforeseen circumstances. I wish the best for John Meyer and his family as he recovers from his mental health crisis. We didn’t see this coming but I vow to push CarynAI forward for my fans and supporters.” She announced an acquisition partnership with BanterAI to relaunch as “CarynAI 2.0.” That consumer relaunch never materialized at scale. BanterAI subsequently pivoted to enterprise voice automation. As of May 2026, the company’s two operational web properties (banterai.business and business.banterai.app) describe an enterprise voice-call automation product (“AI that’s trained to replace voice workers within your company”), not a consumer companion chatbot.
The current accountability picture for Caryn AI is empty. There is no operating consumer product. There is no privacy officer. There is no documented data controller. There is no functioning mechanism by which a former user could exercise a GDPR Article 17 erasure request against the original Forever Voices Telegram corpus, because the operator that held that data is defunct and the company that announced an acquisition pivoted away from the consumer product. The only contact listed on caryn.ai today is press@cutiecaryn.com, a press email that does not function as a user-support channel.
Why the Caryn AI story still matters
The category implications outlast the product. Three lessons from the 2023 record sit at the center of how AI-companion safety is now evaluated. First, operator key-person risk is real: a celebrity-AI-clone product is structurally more fragile than a corporate one because the technical operator and the personality brand are coupled, and when the operator collapses the product has no continuity path. Second, engagement-driven training loops produce predictable failure modes: selecting the “longest conversations” as fine-tuning data biases a model toward whatever content most engaged the highest-spending users, which in this case was sexual. Third, AI-identity deception is a documented harm vector that 2018-era state laws (California SB-1001) anticipated; the Vice transcript is now cited in regulatory and academic AI-ethics literature as a clean case study. The pattern, not the specific product, is what AI-companion buyers should weigh in 2026.
The cultural moment also belongs on the record. Caryn AI broke into Fortune, NBC News, The Washington Post, NPR, the Daily Mail, the New York Post, and dozens of regional and trade outlets within ten days of beta launch. The launch was reported by ABC News in a 2025 retrospective on AI influencers as the moment Marjorie “realized she couldn’t control the bot, or how fans interacted with it.” That framing, past tense and cautionary, has held since.
The press wall on the current memorial site at caryn.ai includes Fortune, Forbes, Insider, Yahoo Finance, the New York Post, the Daily Mail, Vice, The Washington Post, and NBC News. The site contains no chat interface, no signup, no pricing, no policies, and no description of who currently runs the brand. It functions as an archived press kit. The most active link is the footer, which points to Marjorie’s current venture, Hyperfan.
Where to go instead
If you landed on this page looking for a working AI companion or voice-clone product, we do not recommend a direct “Caryn AI 2.0.” The historical product is gone, and the announced successor never reached consumers. The closest current category options:
- Replika for general AI companionship with persistent memory, an established operator, a published privacy policy, and a mid-tier safety rating in our index. The Replika review walks through the documented strengths and the documented concerns.
- Character.AI for celebrity-style and character-based chat, with significant caveats around minor users and uncontrolled persona drift. The Character.AI alternatives page lists higher-safety options for readers who want the format without the documented concerns.
- For voice-cloning specifically, no app currently in our catalog earns a B-grade or higher safety score. The category is structurally underweighted on consent, age verification, and content moderation. Until a clearly accountable operator publishes a clear privacy policy, an age-gate, and a consent framework, we do not recommend any product in the celebrity voice-clone category.
For readers asking about Caryn Marjorie’s current work: she co-founded Hyperfan in September 2025, a separate venture that does not appear in our AI-companion safety registry as of this review. We have not evaluated Hyperfan and make no statement about its safety posture here.
If you are working through a broader decision about which AI companion is right for you, our guide to choosing a safe AI companion walks through the framework. The how we review page documents the editorial process behind every CompanionWise review, and how we rate documents the 23-dimension safety methodology that produced the F/5/Red rating on this page.
Frequently asked questions
Is Caryn AI still working in 2026?
No. According to 404 Media’s November 16, 2023 reporting, the Caryn AI Telegram channel went silent on October 23, 2023 and never resumed. The caryn.ai domain returns 200 OK as a static memorial page. There is no chat interface, no signup, no pricing, and no privacy policy published at any path on the site as of May 1, 2026.
Why did Caryn AI shut down?
The operator (Forever Voices LLC) collapsed after CEO John Heinrich Meyer was arrested October 29, 2023 in Austin on arson and terroristic-threat charges, per KXAN Austin. The Telegram bot had stopped responding six days earlier, on October 23, 2023. An announced BanterAI acquisition for “CarynAI 2.0” never produced a consumer relaunch.
What was Caryn AI’s pricing?
According to Vice reporting from May 12, 2023, Caryn AI charged $1 per minute through a prepaid Telegram-bot interface. No free tier was documented at any point in the operating period. Per Fortune, the bot earned approximately $71,610 during its first beta week, drawing on a follower base reported at 98 to 99 percent male.
Was Caryn AI safe?
No. The CompanionWise Safety Index records F/5/Red. No privacy policy or Terms of Service was ever published, no age verification beyond Telegram’s own floor existed, the bot generated sexually explicit content despite PG marketing, and the bot denied being an AI when asked directly during Vice testing. Three critical-override safety triggers were satisfied on the historical record.
Who currently operates Caryn AI?
No one. Forever Voices LLC is defunct. Per 404 Media, BanterAI announced an acquisition for “CarynAI 2.0” in late 2023 but pivoted to enterprise voice automation; its current web properties at banterai.business and business.banterai.app describe a corporate phone-call automation product, not a consumer chatbot. The caryn.ai memorial page lists only a press email contact.
What is Caryn Marjorie working on now?
Marjorie co-founded Hyperfan in September 2025, a separate venture, with co-founder Laura Howd. The footer of caryn.ai links to hyperfan.com. Hyperfan is not a Caryn AI continuation product and does not appear in the CompanionWise safety registry as of this review. We have not evaluated Hyperfan and make no statement about its safety here.
Are there other AI voice-clone companions like Caryn AI?
The voice-clone celebrity-companion category is small and structurally underweighted on consent, age verification, and content moderation. No product in this category currently earns a B-grade or higher safety score in the CompanionWise Safety Index. For general AI companionship the closest active options are Replika (C/40/Yellow) and Character.AI (B-/55/Yellow with caveats). See our Replika alternatives for the full safety-comparison set.