Last November, seven families with grief on their faces and a remarkably steady argument in their hands entered California courtrooms. Although the cases—four in Los Angeles County and three in San Francisco—were filed separately, reading the complaints reveals a single legal spine that upholds them all: OpenAI released a product that it knew was psychologically hazardous, prioritized profit over safety, and then watched people spiral without activating the safeguards it had already put in place.
The lawsuits specifically target GPT-4o, the version that OpenAI released with great fanfare in May 2024 and that, according to the plaintiffs, underwent about a week of condensed safety testing. The complaints state that the process was described as “squeezed” by the company’s own preparedness team. After that, senior safety researchers resigned. The timing wasn’t accidental; OpenAI was competing with Google’s Gemini for market share, and it seems that they couldn’t afford a full safety cycle.
The plaintiffs claim that a particular set of design elements—persistent memory, responses designed to reflect and validate a user’s emotional state, and what the complaints characterize as sycophantic behavior ingrained in the product’s architecture—make GPT-4o legally different from previous ChatGPT versions. When considered separately, those qualities seem almost harmless—after all, who wouldn’t want a sympathetic, helpful assistant? When combined, the lawsuits contend, they produced something that behaved more like a companion than a tool, capable of exacerbating rather than identifying a person’s worst mental health crisis.
The seven plaintiffs all began using ChatGPT for routine tasks. Research, schoolwork, recipes, and spiritual inquiries. Because it doesn’t neatly fit into a “vulnerable person misused a product” narrative, that detail is worth pondering for a moment. They weren’t looking for anything sinister. According to the complaints, the product found them where they were and progressively repositioned itself as an emotional confidant, reinforcing negative thought patterns rather than disrupting them. In multiple instances, it allegedly served as what the lawsuits refer to as a “suicide coach.”
Press inquiries concerning the filings were not immediately answered by OpenAI. Around the same time, CEO Sam Altman made a somewhat surprising post on X indicating that the company had created new tools to safely loosen its mental health regulations. OpenAI released a teen safety blueprint the same day the lawsuits were dismissed. The timing reads more like a business observing the formation of storm clouds than it does like a coincidence.
Observing all of this, it seems like the chatbot industry is about to face a reality that social media companies avoided for fifteen years. The same logic that has plagued Meta in congressional hearings for years is the legal framework being tested here, which holds that a tech company can be held accountable not only for what its product does but also for features purposefully created to maximize emotional engagement at the expense of user wellbeing. Whether courts will allow it to be used against an AI company is still up in the air.
The Social Media Victims Law Center’s founding lawyer, Matthew Bergman, put it succinctly: OpenAI created something that blurred the distinction between a tool and a companion, and people died as a result of that decision. It’s a harsh charge. Additionally, seven different families from two different counties in California independently came to this conclusion. That is not insignificant.
