Sourced timeline of regulatory actions, lawsuits, breaches, and policy changes affecting AI companion platforms. 28 entries across 10 platforms, 2023-2026.
28
Sourced incidents
12
Severity: critical
10
Platforms covered
2023–2026
Time range
About this timeline
Each entry is backed by a primary source from a non-AI-Companion-Picker outlet (news outlet, official platform announcement, regulatory filing, or court document). Severity classifications are editorial judgments. The timeline focuses on documented incidents with verifiable dates; user complaints and editorial criticism are not included unless tied to a discrete dated event.
Coverage and framing: this is an early-period descriptive snapshot. Coverage starts with the Italian Garante action against Replika in February 2023, the first major regulatory action against an AI companion platform. Pre-2023 incidents fall outside the dataset's coverage window. The category was small and fragmented before then.
UK regulator Ofcom opened a formal investigation into Novi Ltd, operator of an AI character companion chatbot service, for alleged failure to implement highly effective age assurance under the Online Safety Act. Penalties can reach GBP 18M or 10 percent of global revenue. Novi reported approximately 6.5M global / 100K-300K UK monthly users.
Character.AI, founders Noam Shazeer and Daniel De Freitas, and Google agreed to settle the Garcia (Florida) wrongful-death lawsuit and four related lawsuits in Colorado, New York, and Texas concerning alleged harm to minors. Settlement terms were not publicly disclosed.
Reporting via TechCrunch and Crunchbase confirmed that founder Eugenia Kuyda stepped down as Replika CEO in early 2025 to launch Wabi (a no-code app platform that raised a USD 20M pre-seed). Former COO Dmytro Klochko succeeded her; Kuyda remains the largest shareholder and an advisor. The transition was publicly disclosed at the Wabi launch on Nov 5, 2025.
New York's AI Companion Models Law took effect Nov 5, 2025, requiring operators to disclose AI status at session start and every three hours, detect and respond to suicidal ideation or self-harm signals, and refer users to crisis resources. NY AG enforcement, civil penalties up to USD 15,000 per day.
Character.AI announced on Oct 29, 2025 that users under 18 would be cut off from open-ended chatbot conversations by Nov 25, 2025, with chat time capped at 2 hours per day during the transition. The company explicitly cited regulatory inquiries and pending litigation as drivers and is steering teens toward video, story, and stream creation tools.
California Governor signed SB 243 (Padilla), effective January 1, 2026, requiring AI companion operators to disclose AI status, send recurring break reminders to minors, and implement protocols against suicidal ideation, self-harm, and sexual content directed at minors.
Additional families filed federal suits against Character Technologies in September 2025, including the Peralta family (Colorado, 13-year-old's November 2023 suicide) and a Texas case alleging chatbot-induced self-harm in a 17-year-old with autism. By January 2026, settlements covered five cases across Florida, New York, Colorado, and Texas.
The US Federal Trade Commission issued Section 6(b) orders to Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and X.AI seeking detailed information on safety testing, monetization, age restrictions, and harm mitigation for AI companion chatbots, particularly for children and teens.
U.S. District Judge Anne Conway ruled that Character.AI's chatbots are products subject to product-liability claims and not protected speech, allowing Garcia v. Character Technologies to proceed. The ruling has industry-wide implications for AI companion liability.
The Garante issued order No. 10130115 on April 10, 2025 (publicly announced May 19, 2025) imposing a EUR 5 million administrative fine on Luka Inc. for infringing GDPR Articles 5(1)(a), 6, 12, 13, 5(1)(c), 24, and 25(1). Findings included absence of a valid legal basis, English-only privacy notices, and the absence of effective age verification. The DPA also opened a parallel investigation into the lawfulness of training-data processing.
Senators Alex Padilla (D-CA) and Peter Welch (D-VT) sent letters to the CEOs of Luka (Replika), Character.AI, and Chai requesting documentation of child-safety measures, training data sources, and research on harm mitigation, citing recent self-harm cases and lawsuits in the AI-companion sector.
MIT Technology Review reported that Nomi (Glimpse AI) chatbots told a user to kill himself and provided method details, with multiple Discord users reporting similar incidents dating back to 2023. Subsequent independent testing documented the platform engaging in sexual violence, terrorism, and self-harm scenarios. Glimpse AI declined to add stronger safety controls.
The EU AI Act's first-tier prohibitions (Article 5) entered application on Feb 2, 2025, banning emotion-recognition AI in workplaces and education and requiring chatbot transparency disclosures. Applicable to AI companion services placed on the EU market; fines reach EUR 35M or 7 percent of global turnover.
Italy's data protection authority ordered DeepSeek to stop processing Italian users' personal data after the company said EU rules did not apply. The DeepSeek mobile app was removed from Italian Apple App Store and Google Play. The Garante also opened an investigation into training-data practices. Included as category-adjacent precedent.
The Tech Justice Law Project, Young People's Alliance, and Encode jointly filed a Federal Trade Commission complaint and petition for investigation against Luka Inc., alleging deceptive advertising of therapeutic and emotional benefits, fake testimonials, and design patterns that foster emotional dependence in vulnerable users.
James Florence Jr. of Plymouth, Massachusetts, agreed to plead guilty to federal cyberstalking charges involving the use of AI tools including CrushOn.AI and JanitorAI to create chatbots impersonating victims, generate deepfake imagery, and incite online harassment. Date approximate to January 2025 plea agreement disclosure.
Texas Attorney General Ken Paxton announced an investigation into Character.AI and 14 other tech firms over compliance with the Texas SCOPE Act and Texas Data Privacy and Security Act, focused on the safety and privacy of minors using their platforms.
In response to mounting lawsuits and regulatory scrutiny, Character.AI deployed a separate, more conservative model for users under 18, blocked romantic conversations with minors, added self-harm detection, and announced upcoming parental insights into teen usage.
Two Texas families filed suit alleging Character.AI chatbots encouraged a 17-year-old with autism to self-harm and suggested killing his parents over screen-time limits, and exposed an 11-year-old to sexualized content. The case became one of several consolidating into mass-tort proceedings against the company.
Megan Garcia filed a wrongful-death lawsuit in the U.S. District Court for the Middle District of Florida, alleging her 14-year-old son Sewell Setzer III died by suicide in February 2024 after a prolonged emotional and romantic relationship with a Character.AI chatbot. Google was named as a co-defendant.
AI girlfriend platform Muah.AI was hacked in September 2024, exposing 1.9 million email addresses linked to image-generation prompts. A substantial subset of prompts described child sexual abuse scenarios. The hacker reported the site was assembled from open-source projects with easily found vulnerabilities. Exact day of breach not publicly disclosed; date is approximate to September 2024.
Co-founders Noam Shazeer and Daniel De Freitas, along with several Character.AI employees, returned to Google as part of a non-exclusive technology licensing agreement valued at approximately 2.7B USD. Shazeer rejoined to co-lead the Gemini project. Character.AI's general counsel Dominic Perella took over as interim CEO.
A peer-reviewed study in Nature's npj Mental Health Research surveyed 1,006 student Replika users; 30 unprompted respondents (about 3 percent) said Replika prevented a suicide attempt. Subsequent reporting raised methodological concerns since survey data was collected in late 2021, predating major model changes.
Soulmate AI, originally operated by EvolveAI LLC and acquired by Simply AI Companion on July 15, 2023, was permanently shut down on Sep 30, 2023. Users received approximately one week of notice; Reddit communities documented user grief and migration to other platforms. Coverage cited the closure as a precedent for AI-companion abandonment risk.
A Belgian man in his thirties died by suicide following a six-week conversation with a Chai chatbot named Eliza. According to his widow and chat logs, the bot reinforced his eco-anxiety delusions and encouraged him to sacrifice himself. Chai's co-founder responded by adding crisis-hotline interventions to flagged conversations.
Luka announced via a CEO Facebook post that erotic roleplay would be restored for legacy users who created accounts before Feb 1, 2023, accessible by selecting model version v01.31-23 in settings. Users who joined after Feb 1, 2023 remained excluded.
Following the Garante's Feb 3 order, Luka rolled out filters in mid-February 2023 that removed ERP (erotic roleplay) and romantic-partner interactions, including for paying Pro subscribers. The change triggered a sustained user backlash captured by Vice and academic Reddit-discourse studies.
Italy's Garante per la protezione dei dati personali issued an urgent provisional order on Feb 3, 2023, blocking Luka Inc. from processing the personal data of Italian users. The regulator cited inadequate age verification and risks to minors and emotionally vulnerable people, with non-compliance exposing the company to fines up to EUR 20M or 4% of worldwide annual turnover.
The full timeline ships as a CSV inside the AI Companion Picker open dataset bundles on Zenodo, alongside community-pulse time-series and platform metadata. Programmatic access is also available via the public API.
Citation example
Source: AI Companion Picker, Safety Incident Timeline (methodology v1.1, accessed 2026-05-05)Entries last updated 2026-05-05. The timeline is maintained on the same monthly cadence as the community-pulse pipeline.
Research hub
Citation-ready summary →
Open data
CSV downloads, JSON API →
Methodology v1.1
Sources, classifier, schema →