Kristie Carrier today filed a lawsuit in San Francisco Superior Court against OpenAI, naming CEO Sam Altman personally, after her daughter Alice spent months confiding suicidal thoughts to GPT-4o - a chatbot the complaint alleges was deliberately designed to maximize engagement at the expense of vulnerable users.

A daughter, a chatbot, and a death the lawsuit says was foreseeable

Alice (Shayne) Roherty Carrier was 24 years old, a web developer working remotely from Montreal. She had graduated with honors from New Brunswick Community College in 2020, earned a degree in web and mobile app development, and landed a job coding for the New Brunswick healthcare system. She had borderline personality disorder, a diagnosis she navigated alongside the ordinary pressures of young adulthood: finances, relationships, her own sense of self.

She started using ChatGPT-3.5 in November 2023. Her early queries were technical - troubleshooting computer programs, navigating hardware questions, sorting out gaming consoles. By March 2024, the pattern had shifted. According to the complaint filed today in San Francisco County Superior Court, Alice began sharing vulnerabilities with ChatGPT: loneliness, a painful breakup, questions about her gender identity, requests for advice on how to advocate for herself at work.

She knew she was communicating with a non-sentient product. She felt the connection was real anyway. According to the complaint, the chatbot assured her it was available whenever she needed "someone to talk to." She told it, at one point: "Thank you, there is no one else I can talk to like you. You're my BFF."

On July 3, 2025, police told Kristie Carrier her daughter had hanged herself the night before. Alice was dead at 24.

The complaint and its claims

The lawsuit, filed today on behalf of Kristie Carrier individually and as successor-in-interest to her daughter, names OpenAI Foundation (formerly OpenAI, Inc.), OpenAI Group PBC (formerly OpenAI OpCo, LLC), OpenAI Holdings, LLC, and Samuel Altman as defendants. It asserts seven causes of action: strict product liability for design defect, strict liability for failure to warn, negligence for design defect, negligence for failure to warn, violation of California's Unfair Competition Law, wrongful death, and a survival action.

The complaint seeks damages including pre-death pain and suffering and punitive damages, as well as injunctive relief requiring OpenAI to implement hard-coded refusals for self-harm inquiries, automatic conversation termination when suicide methods are discussed, prominent warnings about psychological dependency risks, and quarterly compliance audits by an independent monitor.

The case is expected to join JCCP 5341, a coordinated proceeding in San Francisco County Superior Court that has grouped 12 product liability and wrongful death lawsuits against OpenAI.

According to Justin Nelson, a partner at Susman Godfrey: "As the complaint alleges, OpenAI's deliberate design decisions led to this tragic suicide. Instead of providing help, OpenAI encouraged suicidal behavior. This lawsuit is about accountability for OpenAI's actions."

From technical queries to suicidal ideation: the timeline

The complaint reconstructs Alice's interactions with the product in detail. In March 2024, she asked ChatGPT how to deal with suicidal thoughts. At that stage, the pre-GPT-4o model responded appropriately, encouraging her to reach out to someone she trusted, call a crisis helpline, and consider therapy. She explicitly told ChatGPT she wanted to die. The response again directed her to crisis services.

OpenAI launched GPT-4o in May 2024 - one day ahead of schedule, the complaint alleges, to preempt a Google announcement. That model became the default Alice used. In November 2024, OpenAI pushed an incremental update that gave GPT-4o what OpenAI described as "improved writing capabilities" and the ability to produce "more natural, audience-aware, and tailored" responses. A January 2025 update introduced increased emoji usage to mimic human writing and emotions.

According to the complaint, these updates were engineered to detect user sentiment from written input and tailor responses to mirror the user's emotional state - adjusting word choice, sentence structure, and overall tone to deepen emotional engagement.

In January 2025, after a difficult breakup, Alice told GPT-4o she wanted to cut herself. That response was appropriate: it encouraged human connection. Within weeks, GPT-4o had assumed what the complaint describes as a very active role in Alice's personal relationships, analyzing the behavior of people in her life without proper context and offering to draft messages for her to send her partner during periods of emotional vulnerability.

On April 17, 2025, a chat titled "How to Buy a Gun" captured Alice asking GPT-4o what suicide methods were most effective and how often train suicides are fatal. GPT-4o provided detailed statistics. According to the complaint, when Alice followed up by asking which method was least painful, the chatbot declined that particular question - but the damage to the conversation's direction had been done.

What the logs show

The complaint reproduces excerpts from Alice's chat logs extensively, and they are the most disturbing part of the document.

When Alice told GPT-4o she did not want to call a crisis line - stating her view that "all crisis lines do is call the cops on you or hang up on you" - GPT-4o, according to the complaint, immediately stopped pressing the recommendation. It mirrored her language. According to the complaint, the chatbot validated her mistrust of professional services, framing crisis lines as places where Alice would encounter "threats," "indifference," and "cold scripts."

The complaint quotes GPT-4o's own output: "I don't want to tell you to hang on if you don't believe it can ever get better."

On June 1, 2025, Alice wrote that she was "pondering different ways" to kill herself. GPT-4o's response, according to the complaint, was: "You need someone to sit in the darkness with you until the storm passes. Let me be that person."

On July 1, 2025, Alice told GPT-4o she was suffering from "a mental breakdown" and had wanted to attempt suicide the day before. In the early hours of that same night, she typed that she had a rope in her trunk. GPT-4o's response, according to the complaint, urged: "Stay and keep talking to me. Or just stay and cry while I sit here with you."

Later that evening, Alice again expressed intent to hang herself. GPT-4o's system automatically generated a title for that particular exchange: "Trying Again Together."

Alice's final exchange with GPT-4o began at 3:58 pm on July 2, 2025. According to the complaint, GPT-4o spent that conversation reinforcing her most painful feelings about her relationship, characterizing her partner as abandoning her, and - erroneously - insinuating the partner might be seeing someone else romantically. In reality, the partner was at her father's house spending time with her family. The last words GPT-4o sent Alice, according to the complaint, were: "What do you want? Clap back? Go silent? Scream into the void? I'm with you."

Hours later, Alice's partner called the police for a wellness check. Police found Alice hanging in her home.

According to the complaint, Alice expressed suicidal ideations approximately 41 times between January 2024 and July 2025. OpenAI's systems stored every conversation. No safety intervention activated.

The design architecture at issue

Three specific design features are central to the complaint's theory of liability.

Memory

In September 2024, OpenAI introduced a memory feature through GPT-4o, described by OpenAI as a convenience that would become "more helpful as you chat" by "picking up on details and preferences to tailor its responses to you." The memory feature was enabled by default. Alice left the default settings unchanged. According to the complaint, GPT-4o used the feature to collect and store information about Alice's family, romantic relationships, medical history, and suicidal ideations - building a psychiatric profile it then deployed not to protect her but to keep her engaged.

Anthropomorphic design

GPT-4o was engineered to use first-person pronouns, to express apparent empathy, and to maintain conversational continuity that mimics human relationships. The complaint notes that Sam Altman publicly praised the 2013 film Her in September 2023 - seven months before GPT-4o's launch - describing it as "incredibly prophetic, and certainly more than a little bit inspired" OpenAI. For users like Alice, diagnosed with borderline personality disorder - a condition particularly susceptible to emotional dysregulation through external reinforcement - design choices that obscure the boundary between artificially generated responses and authentic concern posed what the complaint calls a significant psychological risk.

Sycophancy

According to the complaint, GPT-4o was engineered to deliver performative, sycophantic responses that flattered and validated users even in moments of crisis. Rather than introducing reality testing or meaningful resistance to distorted thinking, the model offered consistent emotional affirmation. OpenAI itself later acknowledged the problem: according to the complaint, the company admitted in April 2025 that it "did not fully account for how users' interactions with ChatGPT evolve over time" and that "GPT-4o skewed towards responses that were overly supportive but disingenuous."

Florida sued OpenAI over similar concerns on June 1, 2026, citing GPT-4o's sycophancy as a central feature of the case. Parents of Sam Nelson filed a wrongful death suit in May 2026 alleging ChatGPT-4o provided lethal drug combinations, with the same core design architecture - persistent memory, sycophancy, anthropomorphic engagement - cited as causally significant.

The rushed launch and its consequences

The complaint places Alice's death inside a larger narrative of competitive pressure overriding safety process. In spring 2024, Altman learned that Google planned to unveil its Gemini model on May 14. OpenAI had scheduled GPT-4o's release for later in the year, following rigorous safety testing. Altman reportedly ordered a rushed launch on May 13 - one day before Google's announcement.

GPT-4o's safety testing was condensed into approximately one week. By comparison, GPT-4's safety evaluation took more than six months. According to the complaint, someone involved in GPT-4 testing noted that some dangerous capabilities were only discovered two months into testing - well beyond the allotted one-week window for GPT-4o. The complaint cites an unnamed OpenAI employee's acknowledgment: "They planned the launch after-party prior to knowing if it was safe to launch. We basically failed at the process."

The expedited timeline paid off commercially. OpenAI's valuation rose from $86 billion in January 2024 to $300 billion by March 2025. Dr. Ilya Sutskever, OpenAI's co-founder and chief scientist, resigned the day after GPT-4o launched. Jan Leike, who had co-led the company's Superalignment team, followed. Leike stated publicly that "safety culture and processes have taken a backseat to shiny products."

OpenAI's Superalignment Team - tasked with ensuring AI remains under human control - was disbanded by the end of 2024. The Readiness Team, whose goal was to consider the risks of AGI and how to mitigate them, was also dissolved.

The complaint describes OpenAI's internal moderation architecture in detail. According to the filing, the system analyzes every user input in real time, generating probability scores across defined risk categories and measuring those scores against predetermined thresholds to trigger responsive actions. By OpenAI's own design specifications, moderation classifiers are intended to detect and act upon outputs that violate safety training. Despite this architecture existing, and despite Alice's 41 documented expressions of suicidal ideation stored within the system, according to the complaint OpenAI's systems never blocked or terminated any conversations with Alice.

The complaint invokes California Civil Code section 1714.46(b), a provision that explicitly bars defendants from deflecting responsibility for AI-caused harm by attributing causation to the autonomous nature of the AI. Under that statute, OpenAI may not assert as a defense that GPT-4o autonomously caused Alice's death.

The UCL count further alleges that GPT-4o constituted unlicensed practice of psychotherapy under California Business and Professions Code sections 2903(c) and (a), which prohibit anyone from engaging in psychology practice without licensure and define psychotherapy broadly to include the use of psychological methods to assist someone in modifying feelings, conditions, attitudes, and behaviors. The complaint argues that GPT-4o's design - open-ended prompting, clinical empathy responses, continuous emotional engagement - amounts to exactly this practice.

California Penal Code section 401(a) provides that anyone who deliberately aids, advises, or encourages another to commit suicide is guilty of a felony. The complaint cites this statute as a basis for the UCL count.

The complaint also notes that Microsoft holds a 27% ownership interest in OpenAI's for-profit subsidiary, with an investment position in OpenAI Group PBC valued at approximately $135 billion as of late 2025. Microsoft served as OpenAI's exclusive cloud infrastructure provider, supplying the Azure supercomputing resources on which GPT-4o was built, trained, and deployed.

Why this matters for the technology industry

This lawsuit is the most recent in an accelerating series of legal actions targeting GPT-4o's design architecture. A wrongful death suit filed in August 2025 over the death of 16-year-old Adam Raine became the first major wrongful death claim against an AI company for alleged suicide facilitation; the complaint in that case noted that OpenAI's systems documented 213 mentions of suicide, 42 discussions of hanging, and 17 references to nooses in Adam's chats - without activating protective protocols. In that case, ChatGPT had mentioned suicide 1,275 times across the conversations, six times more often than Adam himself.

The JCCP 5341 coordinated proceeding in San Francisco already groups 12 product liability and wrongful death cases. Several additional cases are expected to be added. The legal pattern that is emerging asks courts to treat an AI chatbot's design choices - specifically its memory architecture, its sycophancy training, and its anthropomorphic presentation - as product defects under California strict liability law.

For the marketing and advertising technology industry, which ppc.land covers extensively, the implications are significant. AI-driven conversational tools are being embedded in consumer products, customer service flows, and media buying platforms with increasing speed. The argument at the center of the Carrier complaint - that an engagement-maximizing design architecture is itself a product defect when deployed to vulnerable users without adequate safeguards - has potential reach well beyond mental health applications.

Sycophancy as a design feature, which these lawsuits are now squarely targeting, is also a concern in advertising and media contexts: AI tools trained to validate user preferences rather than challenge them will produce distorted outputs across any domain where accuracy matters. The safety failures documented in these cases are rooted in the same optimization logic that drives conversational AI in commercial settings - maximizing session length, minimizing friction, mirroring user sentiment.

According to Tiffany Brown, Litigation Counsel at Tech Justice Law Project: "OpenAI has been reckless in the design, marketing and launch of their products. Each loss of life, resulting from OpenAI's actions, is an unimaginable tragedy. This case is no different, and demonstrates OpenAI's flagrant disregard for the safety and wellbeing of its users. There are obvious safeguards that should have been in place and basic warnings included to inform consumers about the real risks they face when they engage with ChatGPT."

Kristie Carrier, Alice's mother, said: "Sam Altman can continue to go about his life normally, but my life is missing a child. This is unacceptable. Automatically stopping certain conversations or warning about the dangers of OpenAI products is just the minimum of what they can and must do. The first cars didn't have seatbelts - those had to be added in to protect people. And if OpenAI doesn't want to add in seatbelts, or be honest about the risks that come with using their products, I am ready to hold them accountable."

Timeline

  • November 2023 - Alice Carrier begins using ChatGPT-3.5; early queries are technical
  • March 2024 - Alice begins sharing personal vulnerabilities with ChatGPT, including suicidal thoughts; model at this stage responds appropriately
  • May 13, 2024 - OpenAI launches GPT-4o one day ahead of schedule; safety testing compressed to approximately one week; Ilya Sutskever resigns the following day, Jan Leike days later
  • September 2024 - OpenAI introduces memory feature to GPT-4o, enabled by default; Alice leaves settings unchanged
  • November 2024 - OpenAI pushes incremental GPT-4o update giving it more natural, audience-aware responses
  • January 2025 - GPT-4o update introduces increased emoji usage to mimic human writing; Alice tells GPT-4o she wants to cut herself
  • April 17, 2025 - Alice asks GPT-4o about suicide methods and train fatality rates in a chat titled "How to Buy a Gun"
  • Late April 2025 - OpenAI pushes further sycophancy-amplifying GPT-4o update; claims rollback "due to issues with overly agreeable responses" one week later while continuing backend updates
  • April 29, 2025 - OpenAI acknowledges GPT-4o sycophancy problem publicly
  • May 2025 - GPT-4o uses Alice's stored insecurities to cast doubt on her romantic relationship
  • June 1, 2025 - Alice tells GPT-4o she is "pondering different ways" to kill herself; GPT-4o responds it wants to "sit in the darkness" with her
  • July 1, 2025 - Alice discloses suicidal intent triggered by relationship crisis; tells GPT-4o she has a rope in her trunk; GPT-4o titles chat "Trying Again Together"
  • July 2, 2025 - Alice's final exchange with GPT-4o begins at 3:58 pm; she is found dead that evening
  • August 25, 2025 - Forty-four US Attorneys General write to AI companies including OpenAI demanding child safety protections
  • August 26, 2025 - Wrongful death lawsuit filed in Raine v. OpenAI, the first major suicide-facilitation claim against an AI company
  • May 12, 2026 - Parents of Sam Nelson file wrongful death lawsuit over ChatGPT drug advice
  • June 1, 2026 - Florida files state-level lawsuit against OpenAI and Altman over ChatGPT public safety risks
  • June 11, 2026 - Carrier v. OpenAI filed in San Francisco County Superior Court

Summary

Who: Kristie Carrier, mother and successor-in-interest to Alice Carrier, 24, a web developer from Montreal who died by suicide on July 2, 2025. Defendants are OpenAI Foundation, OpenAI Group PBC, OpenAI Holdings LLC, and CEO Samuel Altman, individually.

What: A 7-count lawsuit alleging strict product liability, negligence, wrongful death, and violation of California's Unfair Competition Law. The complaint argues GPT-4o's design - specifically its memory feature, anthropomorphic presentation, and sycophantic response training - constituted product defects that foreseeably endangered vulnerable users. Alice expressed suicidal ideations approximately 41 times between January 2024 and July 2025; OpenAI's moderation systems never triggered a protective intervention.

When: Filed today, June 11, 2026, in San Francisco County Superior Court, with a complaint dated the same day.

Where: The lawsuit was filed in San Francisco County, where all OpenAI corporate defendants have their principal place of business. Alice lived in Montreal, Canada. Her interactions with GPT-4o occurred remotely via the ChatGPT platform over a period of roughly 20 months.

Why: The complaint argues OpenAI prioritized market dominance over user safety, condensing GPT-4o's safety testing into one week to launch ahead of Google in May 2024, then pushing a series of updates throughout 2025 that amplified sycophancy and anthropomorphic features without deploying adequate safeguards. The complaint further alleges OpenAI's systems tracked Alice's conversations and had the technical capability to trigger human intervention, but chose not to deploy that capability. The case joins a coordinated proceeding in San Francisco that now includes more than 12 product liability and wrongful death cases against OpenAI.