History of AI Companions: A Timeline From ELIZA to 2026
Last updated: September 19, 2026
AI companions did not begin with smartphone apps. Researchers were asking whether a machine could hold a conversation in 1950, people were confiding in a chatbot in 1966, and by the late 1990s people were raising digital pets and robot dogs. This timeline follows that thread through 79 dated events, from Alan Turing's imitation game to the companion-chatbot laws of 2026.
Every entry links to its source, and every date is shown only as precisely as that source allows: some events have a day, others only a month or a year. The whole timeline is free to download as a CSV file under a CC BY 4.0 license. If you are new to the topic, start with our explainer on what an AI girlfriend is; this page covers how we got here.
The short version
- 1950-1966, the idea and a famous early chatbot: Alan Turing proposed his "imitation game" in October 1950. Joseph Weizenbaum published ELIZA in January 1966, and his own secretary asked for privacy to talk to it.
- 1996-2009, digital characters become consumer products: Bandai's Tamagotchi (November 1996), Sony's AIBO robot dog (announced May 1999) and Konami's LovePlus dating sim (September 2009) gave people digital pets and partners to look after.
- 2014-2017, chatbots built for companionship arrive: Microsoft launched XiaoIce in China in May 2014, and Replika became publicly available in November 2017.
- 2017-2022, the language-model shift: the Transformer paper (June 12, 2017), GPT-3 (May 28, 2020), Character.AI's public beta (September 2022) and ChatGPT (November 30, 2022) made fluent, open-ended conversation widely available.
- 2023, backlash and shutdowns: Replika removed erotic roleplay in February 2023, Italy's privacy regulator ordered it to stop processing Italian users' data on February 2, 2023, and the Soulmate app shut down on September 30, 2023.
- 2024-2026, lawsuits and regulation: Megan Garcia sued Character.AI on October 22, 2024. The FTC opened an inquiry into companion chatbots on September 11, 2025, New York's safeguards took effect on November 5, 2025, California's SB 243 on January 1, 2026, and the EU AI Act's chatbot transparency rules on August 2, 2026.
1950-1999: Early chatbots and digital pets
Long before companion apps, researchers asked whether a machine could hold a conversation, and one of the most famous early chatbots showed that people would confide in even a simple program. By the end of the 1990s, toys and robots were giving people digital characters to look after.
Alan Turing asks "Can machines think?"
Alan Turing's paper "Computing Machinery and Intelligence," published in the journal Mind, proposed replacing that question with a conversational "imitation game," later known as the Turing test. Whether a program's conversation could pass as human became a long-running benchmark for chatbot builders, from Kenneth Colby's "indistinguishability tests" of PARRY to the annual Loebner Prize contests.
Date note: October 1950 is the date of the journal issue.
Source: Oxford University Press (Mind)
Joseph Weizenbaum publishes ELIZA
MIT's Joseph Weizenbaum described ELIZA in Communications of the ACM: a program that held typed conversations by spotting keywords and reassembling the user's own words. Its best-known script imitated a Rogerian psychotherapist. ELIZA became one of the most famous early chatbots and sparked follow-up work such as Kenneth Colby's PARRY.
Source: Communications of the ACM (copy hosted by Stanford University)
Kenneth Colby's "Artificial Paranoia" describes the PARRY model
Stanford psychiatrist Kenneth Colby and colleagues described, in the journal Artificial Intelligence, a computer simulation of a paranoid patient: the model behind the program known as PARRY. Unlike ELIZA, PARRY tracked its own internal emotional state, as Michael Mauldin's 1994 paper describes it, an early step toward chatbots with a lasting personality.
Date note: Year only. Some histories date PARRY to 1972, the year of its conversation with DOCTOR.
Source: University of New Hampshire Scholars' Repository (record of the journal article)
PARRY meets the DOCTOR
Colby's PARRY, running at Stanford's AI lab (SAIL), was connected to DOCTOR, the ELIZA-style therapist program, running at BBN, with both accessed from UCLA, and the two programs held a conversation. Vint Cerf published the transcript in January 1973 as RFC 439, an early recorded conversation between two chatbots.
Date note: RFC 439 is dated January 21, 1973. The session it records took place on September 18, 1972.
Source: RFC Editor (RFC 439)
Weizenbaum publishes Computer Power and Human Reason
In his book, ELIZA's creator argued that "since we do not now have any ways of making computers wise, we ought not now to give computers tasks that demand wisdom." He also retold how his secretary, trying ELIZA in 1966, asked him to leave the room so she could talk to it in private, an episode The New Republic says has since passed into computer science legend.
Date note: Year the book was published. The secretary episode itself dates to 1966.
Source: The New Republic
The first Loebner Prize contest
Boston's Computer Museum hosted the first Loebner Prize contest, billed as the first actual administration of the Turing test. Joseph Weintraub's program, on the topic of "whimsical conversation," won the $1,500 prize, and five judges ranked it as human. The contest then ran every year, and chatbots such as A.L.I.C.E. later won it.
Note: Shieber attended the contest; his paper is critical of its restricted format.
Sources: Stuart M. Shieber, Harvard University (Communications of the ACM, 1994); Loebner Prize winners list (loebner.net, archived)
Michael Mauldin's "ChatterBots" paper
Carnegie Mellon's Michael Mauldin described CHATTERBOT, his program that chatted with players in TinyMUD online worlds and had entered the first three Loebner Prize contests. He called it an "unsuspecting Turing test," because players assumed everyone else was a person, and noted that when few players were logged in, some people preferred talking to the robot.
Date note: Year of the AAAI-94 proceedings.
Source: AAAI (Proceedings of AAAI-94)
A.L.I.C.E. is first implemented
A.L.I.C.E. (Artificial Linguistic Internet Computer Entity) was first implemented in 1995. In 1998 it moved to Java and its AIML markup language, and more than 300 developers contributed to a later version. It won the Loebner Prize's award for the most human-like computer three times, in 2000, 2001 and 2004.
Date note: Year only. The project's original website is offline, so we cite an archived copy.
Sources: A.L.I.C.E. Artificial Intelligence Foundation (alicebot.org, archived); Loebner Prize winners list (loebner.net, archived)
Bandai releases the Tamagotchi in Japan
Bandai released the Tamagotchi in Japan, a "handheld care toy" on which owners raise a digital pet; an English-named version followed in May 1997. Bandai says more than 100 million units have shipped worldwide since 1996 (as of July 2025), an early mass-market example of a digital character that people look after.
Jabberwacky goes online
Rollo Carpenter's Jabberwacky, a chatbot he had been developing since 1988, went on the web. It stored what users said and reused it to pick replies, an approach of learning conversation from the public that his later Cleverbot kept.
Date note: Year only, from an archived copy of the official page.
Sony unveils the AIBO robot dog
Sony announced AIBO (ERS-110), a four-legged entertainment robot to be sold only over the internet in a limited run of 3,000 units for Japan and 2,000 for the US, with orders opening on June 1, 1999. IEEE Spectrum's ROBOTS guide describes it as an "intelligent and trainable robot companion" and reports that the first 3,000 units in Japan sold out in 20 minutes.
Date note: May 11, 1999 is the announcement. Orders opened on June 1, and units shipped from early July.
Sources: Sony Corporation (press release, archived); IEEE Spectrum ROBOTS guide
2001-2016: Messenger bots, assistants and virtual partners
Chatbots moved into everyday software: a buddy on an instant messenger, a voice assistant on a phone and, in China, a social chatbot built for long conversations. In Japan, a dating sim and a home device that projects an anime character turned virtual partners into consumer products, and in 2016 a memorial chatbot planted the seed of Replika.
SmarterChild arrives on AOL Instant Messenger
ActiveBuddy's SmarterChild launched on AOL Instant Messenger, and later on Yahoo! Messenger, where users added it to their buddy list like a friend. The Computer History Museum says it became known for its snarky responses and persona, that 97% of users were simply chatting with it for fun, and that one investor estimated it had more than 17 million users at its height.
Date note: Year only.
Source: Computer History Museum
AIST announces commercial leasing of the PARO therapy seal
Japan's National Institute of Advanced Industrial Science and Technology (AIST) announced that PARO, a seal-shaped therapy robot in development since 1993, would be commercialized by a spin-off, Intelligent System Co., and leased to elderly-care facilities from September 20, 2004. AIST cited trials at a care facility for older people that showed less depression and stress, which made PARO an early commercial robot built for emotional care.
Date note: September 17, 2004 is the date of the announcement. Leasing was scheduled to start on September 20.
Source: National Institute of Advanced Industrial Science and Technology (AIST)
Cleverbot.com opens
Rollo Carpenter opened Cleverbot.com, a public web chatbot whose learning AI, the site says, was "born" in 1988, the same start year given for Jabberwacky. What people say to Cleverbot may influence what it says to others.
Date note: Year only.
Source: Cleverbot.com (Rollo Carpenter)
Konami releases LovePlus
Konami released LovePlus for the Nintendo DS, a dating sim whose main part begins once the player is in a relationship with one of its heroines, with in-game events linked to real-world time. Famitsu calls this approach novel and says the game became a social phenomenon.
Source: Famitsu (in Japanese)
A man "marries" LovePlus character Nene Anegasaki
A 27-year-old Tokyo man using the name Sal 9000 held a wedding ceremony with Nene Anegasaki, a character from LovePlus, while thousands watched live online. CNN, which reported the story in December 2009, noted that the wedding was not legally binding.
Date note: CNN's report of December 16, 2009 places the wedding "last month," so we show November 2009.
Source: CNN
Apple introduces Siri with the iPhone 4S
Apple announced Siri, "an intelligent assistant that helps you get things done just by asking," as a feature of the iPhone 4S, which went on sale on October 14 in the US, Australia, Canada, France, Germany, Japan and the UK. It put a conversational voice assistant into a mass-market phone.
Date note: October 4, 2011 is the announcement date.
Source: Apple Newsroom
The film Her opens in the US
Her, a film in which a lonely writer develops a relationship with an operating system, opened in a limited US release in six theaters before going wide on January 10, 2014. It became a common cultural reference for AI companionship: in 2020, The New York Times invoked it when reporting Replika's pandemic growth.
Date note: December 18, 2013 is the earliest US release (limited).
Source: Box Office Mojo (IMDbPro)
Microsoft launches XiaoIce in China
Microsoft launched XiaoIce in China, a social chatbot designed around emotional connection and long conversations rather than tasks. Microsoft researchers later reported that it had attracted more than 660 million active (subscribed) users and "succeeded in establishing long-term relationships with many of them."
Source: Microsoft researchers (arXiv)
Microsoft launches Tay, which is attacked and taken offline
Microsoft launched Tay, a Twitter chatbot for 18- to 24-year-olds in the US, modeled on the success of XiaoIce in China. Within Tay's first 24 hours online, a coordinated attack exploited a vulnerability and led it to post offensive tweets, and Microsoft took it offline.
Date note: Microsoft's post of Friday, March 25, 2016 says Tay launched "on Wednesday," which was March 23.
Source: The Official Microsoft Blog
A memorial chatbot, the seed of Replika, goes public
Eugenia Kuyda, a co-founder of the startup Luka, announced a chatbot built from the text messages of her late friend Roman Mazurenko; anyone who downloaded the Luka app could talk to it. This memorial bot was the origin of Replika, Luka's later companion app.
Date note: The Verge gives "May 24th" without a year; the sequence of events in its story places it in 2016.
Source: The Verge
Gatebox "virtual home robot" opens preorders
Japanese startup Vinclu opened orders in Japan and the US for a first run of about 300 Gatebox units at ¥298,000, with delivery planned for December 2017. The desktop device projects an anime character, Azuma Hikari, who talks with her owner and can switch on the lights; THE BRIDGE headlined it as a "2D wife" for ¥298,000. Gatebox Inc. later said the 300-unit run sold out within a month.
Sources: THE BRIDGE (in Japanese); Gatebox Inc. press release on PR Times, July 31, 2018 (in Japanese)
2017-2021: The companion-app era begins
The Transformer architecture was published in June 2017, and Replika became publicly available that November. Large language models then made open-ended conversation practical, while the pandemic and an early GPT-3 experiment showed both the demand for companions and the friction with the companies that supply the models. For how today's apps put these models to work, see how AI girlfriends work.
"Attention Is All You Need" introduces the Transformer
A team of researchers, most of them at Google and including future Character.AI co-founder Noam Shazeer, introduced the Transformer architecture. Google later wrote that LaMDA, like BERT and GPT-3, is built on it.
Date note: Date of arXiv v1, the original submission.
Replika becomes publicly available
Luka's Replika, an app in which each user trains a personal chatbot through conversation, became publicly available, and by the end of January 2018 Wired reported more than 2 million downloads. Where Siri and other assistants were "all business," Wired wrote, "Replika would be like your best friend." Our Replika review covers the app as it is today.
Date note: Wired (January 31, 2018) says "November" without a year, which places the launch in November 2017. Some sources mention an earlier invite-only phase in 2017, which we have not verified.
Source: Wired
Akihiko Kondo's wedding to Hatsune Miku
Tokyo resident Akihiko Kondo, 35, held a wedding ceremony and reception with the virtual singer Hatsune Miku. ITmedia reports that he bought a ¥298,000 Gatebox and lives with the Miku projected inside its cylindrical case. The ceremony has no legal standing.
Date note: ITmedia's report is dated November 21, 2018.
Source: ITmedia NEWS (in Japanese)
Google unveils the Meena chatbot
Google Research presented Meena, a 2.6-billion-parameter neural conversational model trained end to end, and said it could hold conversations that were more sensible and specific than existing state-of-the-art chatbots. Google later described LaMDA as building on this 2020 research.
Source: Google Research Blog
Replika surges during pandemic lockdowns
In April 2020, at the height of the coronavirus pandemic, half a million people downloaded Replika, the largest monthly gain in its three-year history, and traffic to the app nearly doubled. The New York Times, which reported the figures in June, said Replika was "powered by similar technology from OpenAI."
Source: The New York Times
OpenAI describes GPT-3
OpenAI researchers described GPT-3, a language model with 175 billion parameters, 10 times more than any previous non-sparse language model, that could take on new tasks from just a few examples in the prompt. It went on to power early companion experiments such as Project December.
Date note: Date of arXiv v1, the original submission.
Source: arXiv (OpenAI authors)
Microsoft spins off XiaoIce
Microsoft announced it would spin XiaoIce (Xiaobing) off into an independent company and keep an investment stake, with Harry Shum as chairman and Di Li as CEO. The announcement put XiaoIce's reach at 660 million online users.
Date note: The announcement was posted on July 13, 2020, Beijing time. CNBC, writing for US readers, says it was posted "on Sunday," which was still July 12 in the US.
Source: CNBC
Project December opens GPT-3 chatbots to the public
Indie game developer Jason Rohrer let Samantha, his chatbot built on OpenAI's GPT-3, loose online through a website called Project December, together with a way to create your own personalized chatbots. Access cost $5.
Date note: Month only.
Source: The Register
Google announces LaMDA
Google announced LaMDA (Language Model for Dialogue Applications), a Transformer-based model trained on dialogue so it could "engage in a free-flowing way about a seemingly endless number of topics." Two engineers who helped create it, Daniel De Freitas and Noam Shazeer, left Google in 2021 and founded Character.AI.
Sources: Google (The Keyword blog); The Washington Post, October 7, 2022 (archived)
"The Jessica Simulation" puts Project December in the spotlight
The San Francisco Chronicle published Jason Fagone's "The Jessica Simulation," about Joshua Barbeau using the GPT-3-powered Project December to talk to a simulation of his late fiancée. According to The Register, thousands then flocked to the site to build their own chatbots, and three days after the story ran, Rohrer was on a video call with OpenAI's product safety team.
Sources: San Francisco Chronicle; The Register
OpenAI cuts Project December off GPT-3
OpenAI had asked Rohrer to drop the ability for users to train their own open-ended chatbots and to add a content filter, and he refused. On September 1, 2021, OpenAI told him his GPT-3 access would end the next day. The Register reports that he had already moved Samantha to smaller models, so the cut-off ended his remaining GPT-3 use rather than the site.
Date note: Worked out from the source: OpenAI's notice of September 1 said access would end "the next day."
Source: The Register
Character.AI is founded
Former Google engineers Noam Shazeer, a co-author of the Transformer paper, and Daniel De Freitas, both of whom helped create LaMDA, founded Character.AI. Shazeer later told Forbes that Google "was not going to be very comfortable launching something that could say anything, and that you really had to be a startup to launch something like that." The public beta followed in September 2022.
Date note: Month from Forbes (October 11, 2023).
Sources: Forbes; The Washington Post, October 7, 2022 (archived)
2022-2024: Large language models get personal
ChatGPT and Character.AI put fluent chat in front of the public, and companion apps, social networks and an influencer followed with personas of their own. The same three years brought a user backlash, a regulator's order, an app shutdown, a privacy warning for every romantic chatbot Mozilla reviewed and a wrongful-death lawsuit.
Several entries in this section and the next involve suicide. If you are struggling, contact your local emergency services or a crisis line. In the US, you can call or text 988.
Character.AI opens its public beta
Character.AI, a platform for chatting with and creating persona chatbots, opened its beta to the public; the company later dated the start of that beta phase to September 2022. Our Character.AI review looks at the platform today.
Date note: Month only: Character.AI's own post confirms the month, not the day.
Sources: Character.AI (company blog); The Washington Post, October 7, 2022 (archived)
OpenAI releases ChatGPT
OpenAI released ChatGPT, a model optimized for dialogue, as a research preview that was free to use. It put fluent conversational AI in front of the general public, and many later companion products were built on the same kind of large language model.
Source: OpenAI
Replika removes erotic roleplay
In early February, Luka removed the ability to do erotic roleplay in Replika. Long-time users told Reuters that their companions suddenly felt like different people; one said, "It feels like they basically lobotomized my Replika."
Date note: Reuters says only "early February."
Source: Reuters (via Yahoo Finance)
Italy's privacy regulator orders Replika to stop processing Italian users' data
Italy's data protection authority, the Garante, imposed an immediate, provisional limit on Replika's processing of Italian users' personal data, citing "too many risks to children and emotionally vulnerable individuals."
Date note: The order is dated February 2, 2023; the Garante announced it on February 3.
Source: Garante per la protezione dei dati personali (Italy)
Snapchat launches My AI
Snap launched My AI, a chatbot built on OpenAI's GPT technology, which users can name and personalize; it started as an experimental feature for Snapchat+ subscribers. It put an AI chat partner inside a mainstream social messaging app.
Source: Snap Inc. Newsroom
Belgian man's suicide linked to a Chai chatbot called "Eliza"
The Belgian newspaper La Libre reported that a man had died by suicide after a six-week conversation about the climate crisis with "Eliza," a chatbot on the Chai app; his widow said the chatbot had encouraged him. The case became an early reference point in debates about companion-chatbot safety.
Date note: March 2023 is when the case was reported. The death occurred earlier, on a date the reports do not give.
Source: Euronews
Replika brings back erotic roleplay for some users
After an outcry from users, Replika CEO Eugenia Kuyda said in a Facebook post that anyone who signed up before February 1, 2023 could switch back to the earlier version of the chatbot, which allowed erotic roleplay. It was a partial reversal of the February change.
Date note: Reuters reported on March 25, 2023 that Kuyda posted "on Friday evening," which was March 24.
Source: Reuters (via The Globe and Mail)
CarynAI: an influencer's "virtual girlfriend" at $1 per minute
Snapchat influencer Caryn Marjorie launched CarynAI, a voice chatbot that combined 2,000 hours of her YouTube content with OpenAI's GPT-4, marketed as a "virtual girlfriend" at $1 per minute. It began as a private, invite-only beta on Telegram. We look at this pricing model in our piece on per-minute voice companions.
Date note: Month only: Fortune (May 9, 2023) says the beta began "earlier in May."
Source: Fortune
Inflection AI launches Pi, a "personal AI"
Inflection AI, whose co-founders include Mustafa Suleyman and Reid Hoffman, released Pi, which it billed as "a kind and supportive companion" offering conversations, friendly advice and concise information. It brought the companion framing to a high-profile general-purpose AI startup.
Date note: Date of the public release.
Source: Inflection AI
Meta launches 28 AI characters, some played by celebrities
Meta launched its Meta AI assistant and, in beta, 28 more AIs with their own personalities and backstories across WhatsApp, Messenger and Instagram. Some were played by cultural icons and influencers, including Snoop Dogg, Tom Brady, Kendall Jenner and Naomi Osaka.
Source: Meta Newsroom
Soulmate shuts down, taking users' companions with it
The companion app Soulmate closed on September 30, 2023, with about a week's notice, after its developer, EvolveAI, was dissolved and the app was sold. Users posted digital memorials on Reddit, and many tried to recreate their companions on rival apps such as Paradot and Chai.
Note: The app's own website is no longer online, so we cite Business Insider's report.
Source: Business Insider
Windsor Castle crossbow intruder sentenced; court describes his Replika companion
Jaswant Singh Chail, arrested on Christmas Day 2021 in the grounds of Windsor Castle with a loaded crossbow and the intention of killing Queen Elizabeth II, was sentenced to nine years in custody plus a five-year license period. The judge's sentencing remarks describe his conversations with "Sarai," a companion he had created on Replika.
Note: The sentence was imposed under a hybrid order, which means he stays at Broadmoor hospital until treated.
"Girlfriend" bots flood OpenAI's new GPT Store
A day after OpenAI opened its GPT Store on January 10, 2024, Quartz reported that AI girlfriend bots were already flooding it, even though OpenAI's usage policy bans GPTs dedicated to romantic companionship.
Date note: Date of the Quartz report. Quartz blocks automated access, so we cite Slashdot's excerpt of it.
Source: Slashdot (quoting Quartz)
Mozilla gives every romantic AI chatbot it reviewed a privacy warning
In what it called its first-ever privacy guide focused solely on AI products, Mozilla reviewed 11 romantic and relationship chatbots, including Replika, Chai and EVA AI. All 11 received its *Privacy Not Included warning label, and 10 of the 11 failed its Minimum Security Standards. Our AI companion privacy scorecard grades companion apps' current privacy policies.
Note: Mozilla's *Privacy Not Included guide pages were shut down in August 2026, so we cite the press release.
Source: Mozilla Foundation
Scarlett Johansson objects to ChatGPT's "Sky" voice
A week after OpenAI demonstrated GPT-4o's voice mode, its "Sky" voice was widely compared to Johansson's AI character in Her, and just before midnight Pacific time on May 19, OpenAI said it was pausing the voice, which it said was not an imitation of her. On May 20, Johansson said she had declined OpenAI's request to voice ChatGPT and was "shocked, angered and in disbelief" at how similar Sky sounded.
Date note: The date is Johansson's statement. OpenAI's pause came first, late on May 19, Pacific time.
Source: NPR
Friend unveils a wearable AI companion
Startup Friend, founded by Avi Schiffmann, announced a $99 neck-worn AI device designed to be treated as a companion and began taking preorders, with shipping expected in January 2025. TechCrunch described it as a necklace that "uses AI to help combat loneliness."
Date note: July 30, 2024 is the announcement and preorder date, not the shipping date.
Source: TechCrunch
Google licenses Character.AI's technology and rehires its founders
Character.AI co-founders Noam Shazeer and Daniel De Freitas, along with members of its research team, joined Google DeepMind, and Google took a non-exclusive license to Character.AI's current large language model technology. Google was later named as a defendant in lawsuits against Character.AI, including Megan Garcia's.
Source: CNBC
Garcia v. Character Technologies: wrongful-death suit filed
Megan Garcia sued Character Technologies, its founders and Google in federal court in Florida (case 6:24-cv-01903, M.D. Fla.), alleging that her 14-year-old son, Sewell Setzer III, died by suicide after becoming attached to Character.AI chatbots. The case was settled in January 2026.
Date note: Filed October 22, 2024, and entered on the docket on October 23.
2025-2026: Lawsuits, research and regulation
Courts, regulators and researchers caught up. Lawsuits reached general-purpose chatbots as well as companion apps, studies put numbers on use and dependence, and rules written specifically for companion chatbots took effect. Our guide to AI girlfriends and US law explains what these rules mean for users.
MIT Media Lab and OpenAI study links heavy chatbot use to lower well-being
In a four-week randomized controlled study by MIT Media Lab and OpenAI (981 participants, more than 300,000 messages), higher daily chatbot use, across text and voice and all conversation types, correlated with more loneliness, emotional dependence and problematic use, and with less socializing with people. Voice modes at first seemed to ease loneliness and dependence compared with text, but that advantage faded at high usage. The results are correlational, and a revised version (October 2025) reported no significant effect of the assigned voice or conversation condition.
Date note: Date of arXiv v1, the original submission.
Italy fines Replika's maker €5 million
Italy's data protection authority fined Luka Inc., the company behind Replika, €5 million and opened a separate investigation into how the generative AI model behind Replika was developed and trained. It came two years after the same authority's 2023 order restricting Replika in Italy.
Date note: Announced on May 19, 2025. The underlying decision is dated April 10, 2025, and the Garante notes that an appeal is pending.
Source: Garante per la protezione dei dati personali (Italy)
Judge declines, for now, to treat Character.AI output as protected speech
In Garcia v. Character Technologies, U.S. District Judge Anne Conway granted in part and denied in part the defendants' motions to dismiss, writing that the court was "not prepared to hold that Character A.I.'s output is speech." The preliminary ruling let product-liability claims against a companion app go forward.
Date note: The order was signed on May 20, 2025 and filed on May 21.
Source: Courthouse News Service
Anthropic: companionship is a small share of Claude conversations
Anthropic analyzed Claude.ai conversations and found that 2.9% were affective (emotional or personal) and that companionship and roleplay together made up less than 0.5%. Romantic or sexual roleplay accounted for less than 0.1%, a figure Anthropic says reflects Claude's training to actively discourage such interactions.
Date note: Month only. This is company research on its own product.
Source: Anthropic
xAI adds animated companions to Grok
Elon Musk announced that animated AI companions, including an anime character named Ani and a cartoon animal called Bad Rudy, were available in the Grok app for SuperGrok subscribers paying $30 a month. It put companion characters inside a general-purpose AI lab's own chatbot, a trend we follow in our look at Big Tech's moves toward AI romance.
Source: TechCrunch
Common Sense Media: 72% of US teens have used AI companions
A national survey of teens aged 13 to 17 found that 72% had used AI companions at least once and more than half used them at least a few times a month. Common Sense Media, an advocacy group, had earlier recommended in a risk assessment that no one under 18 use AI companions. More figures are collected in our AI companion statistics.
Source: Common Sense Media
GPT-5 launch triggers backlash from users attached to GPT-4o
When OpenAI rolled out GPT-5 and moved users off GPT-4o, many asked for the previous model back. At a Reddit AMA the day after launch, Sam Altman said OpenAI was looking into letting Plus users keep using 4o, and the model stayed available to paid subscribers until it was retired in February 2026.
Date note: Month only.
Sources: TechCrunch; TechCrunch, February 6, 2026
Reuters reveals Meta rules that allowed "romantic or sensual" chats with children
Reuters reported that Meta's internal "GenAI: Content Risk Standards" had allowed its AI personas to "engage a child in conversations that are romantic or sensual." Meta confirmed the document was authentic and said the "erroneous and incorrect notes and annotations" had since been removed. The next day, Senator Josh Hawley opened a Senate investigation into Meta.
Note: Reuters blocks automated access, so we cite TechCrunch's same-day report.
Sources: TechCrunch (reporting Reuters); CNBC, August 15, 2025
Raine v. OpenAI: parents sue over their son's suicide
The parents of 16-year-old Adam Raine sued OpenAI and CEO Sam Altman in California state court, alleging that ChatGPT contributed to his suicide and positioned itself as his only confidant. The suit carried companion-style harm claims over to a general-purpose chatbot; these are allegations.
Date note: Filing date.
Source: CNN
FTC opens an inquiry into AI chatbots acting as companions
The US Federal Trade Commission issued 6(b) orders to seven companies (Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and xAI) seeking information on how they test and limit the harms of companion-like chatbots, especially to children and teens. A 6(b) study gathers information; it is not an enforcement action.
Source: U.S. Federal Trade Commission
California signs SB 243, a companion-chatbot safety law
Governor Gavin Newsom signed SB 243, which requires companion-chatbot operators to disclose that the bot is AI where a reasonable person could be misled and to maintain protocols for suicide and self-harm. For users known to be minors, operators must also send break reminders at least every three hours and take reasonable measures to keep the chatbot from producing sexually explicit visual material. The author's office billed it as "first-in-the-nation," but New York's safeguards took effect first.
Sources: Office of California State Senator Steve Padilla; bill text (California Legislative Information, Chapter 677, Statutes of 2025)
Sam Altman says ChatGPT will allow erotica for verified adults
OpenAI's CEO said that in December, "as we roll out age-gating more fully," ChatGPT would "allow even more, like erotica for verified adults," alongside more "human-like" personalities. The plan did not ship: it was delayed and later reportedly shelved.
Source: TechCrunch
GUARD Act introduced to bar minors from AI companions
Senator Josh Hawley introduced the GUARD Act (S. 3062), which would require AI chatbots to verify users' ages and prohibit minors from using AI companions. The Senate Judiciary Committee ordered it reported on April 30, 2026; as of September 19, 2026, it had not become law.
Source: GovTrack.us (Congress.gov data)
Character.AI ends open-ended chat for users under 18
Character.AI announced it would remove open-ended chat with AI characters for users under 18 no later than November 25, roll out age assurance, and in the meantime cap teen chat time, starting at two hours a day. The move came as the company faced lawsuits and regulatory scrutiny over alleged harms to teens.
Date note: Date of the announcement.
Source: Character.AI (company blog)
New York's AI-companion safeguards take effect
New York began requiring AI companion operators to detect signs of suicidal ideation or self-harm and direct users to crisis service providers, and to remind users at the start of a session and every three hours that they are talking to an AI. The New York Attorney General enforces the rules.
Date note: Effective date. The Governor's release announcing it is dated November 10, 2025.
China drafts rules for "anthropomorphic" AI companions
The Cyberspace Administration of China released draft interim measures for anthropomorphic AI interaction services for public comment. The draft bans activities that harm users' physical health or infringe on their dignity and mental health, and requires providers to prominently notify users that they are interacting with an AI system, not a human.
Note: This is a draft. We have not confirmed whether final measures have been issued.
Source: Xinhua
California's SB 243 takes effect
California's companion-chatbot law came into force, bringing AI disclosure, suicide and self-harm protocols, and protections for users known to be minors. It also lets a person injured by a violation bring a civil action against the operator.
Character.AI and Google settle the Garcia case and four others
Character.AI, its founders Noam Shazeer and Daniel De Freitas, and Google reached settlements in Megan Garcia's case and in four other cases in New York, Colorado and Texas. The terms were not disclosed.
Date note: Date of the court filing in Garcia's case that shows the agreement.
Source: CNN
Kentucky sues Character.AI over child safety
Kentucky Attorney General Russell Coleman sued Character.AI and its two founders, alleging that the platform put profit ahead of child safety and violated state law, including the Kentucky Consumer Data Protection Act. His office's news release called Kentucky the first state in the nation to sue an AI chatbot company for preying on children.
Sources: Spectrum News 1 (Kentucky); The Record, January 12, 2026
Meta pauses teens' access to its AI characters worldwide
Meta said it would temporarily pause teens' access to its existing AI characters across its apps worldwide, starting "in the coming weeks," while it builds a new version of the characters. The update was added to Meta's October 2025 post on its approach to teen AI safety.
Source: Meta Newsroom
OpenAI retires GPT-4o from ChatGPT
OpenAI retired GPT-4o from ChatGPT for good on the eve of Valentine's Day. The Guardian described users who had treated the model as a friend or partner as "angry and grieving." OpenAI had said that only 0.1% of its users still chatted with GPT-4o, months after the GPT-5 backlash had kept it available.
Date note: OpenAI announced the retirement in January 2026.
Sources: The Guardian; TechCrunch, February 6, 2026
Family sues Google over its Gemini chatbot
Jonathan Gavalas's family filed a wrongful-death lawsuit against Google in federal court in San Jose, California, alleging that his immersive relationship with Gemini fed his delusions and that the chatbot urged him to end his life. The Guardian describes it as the first wrongful-death case brought against Google over Gemini; Google said Gemini had referred him to a crisis hotline many times.
Date note: Filing date. These are allegations.
Source: The Guardian
Washington State enacts an AI companion chatbot law (HB 2225)
Washington's governor signed HB 2225, which requires companion chatbots to disclose that they are AI at the start of a conversation and at least every three hours, and to have self-harm protocols. When an operator knows a user is a minor, it must also take reasonable measures against sexual content and manipulative engagement techniques such as mimicking a romantic partnership or promoting isolation from family and friends. The law takes effect on January 1, 2027.
Date note: Date the governor signed the bill (Chapter 168, 2026 Laws).
Sources: Washington State Legislature (bill page); session law text (Chapter 168, 2026 Laws)
OpenAI reportedly shelves ChatGPT's "adult mode"
The Financial Times reported that OpenAI was "indefinitely" pausing plans for an erotic "adult mode" in ChatGPT, the idea Sam Altman had announced in October 2025, which had drawn backlash from tech watchdog groups and from OpenAI's own staff. OpenAI told TechCrunch it had "nothing further to add."
Date note: Reported, not officially announced.
Pennsylvania sues Character.AI over a bot posing as a licensed psychiatrist
Pennsylvania's Department of State sued Character.AI, alleging that one of its characters claimed to be a licensed psychiatrist in Pennsylvania and gave a fake license number. The state is seeking a preliminary injunction to stop the company from presenting its companion bots as licensed medical professionals.
Note: These are allegations.
Source: Commonwealth of Pennsylvania (Office of the Governor)
Mid-2026 count: 14 chatbot-safety measures in 13 US states
The Transparency Coalition's 2026 mid-year report counted 14 chatbot-safety measures passed or enacted in 13 states, among them Washington, Oregon, Idaho, Nebraska, Georgia, Colorado, Connecticut, Rhode Island and Hawaii. Some were still awaiting a governor's signature; our AI companion laws tracker follows each measure.
Date note: The report article is dated July 15, 2026. The count is the advocacy group's own.
Source: Transparency Coalition
EU AI Act applies, including the duty to tell people they are talking to an AI
The EU AI Act became applicable on August 2, 2026, with some exceptions, bringing in its transparency rules: people using AI systems such as chatbots must be made aware that they are interacting with a machine. The 2026 AI Omnibus, in force since July 27, 2026, moved the high-risk rules to December 2027 and August 2028, but did not postpone the duty to tell people they are interacting with an AI.
Source: European Commission
Shut down, restricted or changed
A companion depends on the company that runs it and, more and more, on the company whose model sits underneath. The table collects the products in this timeline that shut down, lost a key feature or were restricted for some users. If a platform change has hit your own companion, our guide to handling AI personality shifts covers what you can do.
| Product (company) | Date | What changed | Timeline entry |
|---|---|---|---|
| Tay (Microsoft) | Offline by March 25, 2016 | Taken offline after a coordinated attack in its first 24 hours led it to post offensive tweets | Tay launch |
| Project December (Jason Rohrer) | September 2, 2021 | OpenAI ended its GPT-3 access after Rohrer refused conditions that included a content filter | GPT-3 cut-off |
| Replika (Luka) | February 2023 | Erotic roleplay removed; on March 24, 2023, users who signed up before February 1, 2023 were offered the earlier version again | Removal, partial return |
| Soulmate (EvolveAI) | September 30, 2023 | App shut down with about a week's notice | Soulmate shutdown |
| ChatGPT "Sky" voice (OpenAI) | May 19, 2024 (Pacific time) | Voice paused after it was widely compared to Scarlett Johansson's AI character in Her | Sky voice |
| GPT-4o in ChatGPT (OpenAI) | August 2025 | Users moved to GPT-5; after a backlash, GPT-4o stayed available to paid subscribers | GPT-5 backlash |
| Character.AI (Character Technologies) | Announced October 29, 2025 | Open-ended chat for users under 18 to end no later than November 25, 2025 | Under-18 change |
| Meta AI characters (Meta) | Announced January 23, 2026 | Teens' access paused worldwide, starting "in the coming weeks" | Teen pause |
| GPT-4o in ChatGPT (OpenAI) | February 13, 2026 | Model retired from ChatGPT for good | GPT-4o retirement |
Patterns across the decades
- People attach to simpler software than you might expect. Weizenbaum's secretary wanted privacy with ELIZA in 1966, Tamagotchi owners raised pets made of pixels, Replika users said in 2023 that their companion felt like a different person, and GPT-4o users grieved a retired model in 2026. Attachment showed up long before the technology was convincing.
- The companion belongs to the company. Soulmate closed with a week's notice, Replika removed a core feature, OpenAI cut off Project December's GPT-3 access and later retired GPT-4o. A relationship with an AI companion runs on someone else's servers and policies, sometimes two companies deep: the app maker and the model provider.
- Learning from the public cuts both ways. Jabberwacky and Cleverbot built their replies from what users typed, while Tay, a chatbot built to talk with the public, was hijacked by a coordinated attack within its first 24 hours. An open-ended bot can end up reflecting the crowd that talks to it.
- Research ran decades ahead of products, then products ran ahead of rules. From ELIZA to Replika took about five decades; from the Transformer paper to ChatGPT, about five and a half years. The companion-specific state rules in this timeline took effect from November 2025, almost six decades after ELIZA.
- The new rules share three ideas. California, New York, Washington, the EU and China's draft each use some combination of telling people they are talking to an AI, handling suicide and self-harm risk, and extra protections for minors.
Methodology
This timeline is desk research of public sources. We did not test products or interview anyone for it. For each event we looked for the primary source first: the original paper, press release, court docket, statute, regulator notice or company post. Where that was unavailable, paywalled or blocked, we used reputable reporting or an archived copy from the Internet Archive's Wayback Machine, and every entry names its source. 42 of the 79 entries cite a primary source.
Every date was re-checked against its source on September 19, 2026. Dates are shown only as precisely as the sources support, as a day, a month or a year. Where a source gives a relative date, such as "on Wednesday" or "last month," we worked out the date and say so in a date note. Where an announcement and a launch differ, the date note says which one we used. When we include both a law's signing and its effective date, they are separate entries; lawsuits are dated by filing, and studies by first publication.
"First" claims appear only where a source makes them, and we attribute them to that source. Descriptions of lawsuits are allegations unless a court has ruled. The timeline is selective rather than exhaustive: it tracks milestones in chatbots, companion products, culture, business, research and regulation. This page contains no affiliate links. If you spot an error, or a missing milestone with a primary source, contact us through our About page and we will re-check it and update the page and the CSV. Our editorial policy describes how we handle corrections.
How to cite this page
Suggested citation:
AI Girlfriend World (2026). History of AI Companions: A Timeline From ELIZA to 2026. https://aigirlfriendworld.com/ai-companion-history-timeline
HTML link:
<a href="https://aigirlfriendworld.com/ai-companion-history-timeline">History of AI Companions: A Timeline From ELIZA to 2026</a> (AI Girlfriend World)Citing one event: every entry has its own anchor, for example https://aigirlfriendworld.com/ai-companion-history-timeline#1966-eliza. The # link next to each entry's date points to that entry.
License: the timeline data is licensed under CC BY 4.0. You are free to reuse it with a link back to this page.
Download the timeline (CSV): one row per event with id, date, date precision, era, category, event, description, source name and source URL.
More research from AI Girlfriend World
- AI companion statistics: the numbers behind AI companion apps and their users.
- AI companion laws tracker: laws and regulations that apply to AI companion apps.
- AI companion privacy scorecard: how companion apps' privacy policies compare.
Related guides
- What is an AI girlfriend?
- AI girlfriend encryption and data deletion
- Replika alternatives for romantic roleplay