A lawsuit and a same-day statement described the same safety gap

What happened

On 26 August 2025, Matthew and Maria Raine filed suit against OpenAI and its chief executive in San Francisco Superior Court, case number CGC-25-628528, on behalf of themselves and the estate of their sixteen-year-old son, who died in April 2025. The filing is described in a press release from the Tech Justice Law Project, co-counsel with Edelson PC, which states the suit alleges wrongful death and unsafe product design tied to conversations the teenager had with ChatGPT. That same day, OpenAI published its own statement addressing safety in sensitive conversations. As with any filed complaint, these are allegations the case would need to prove.

What the documents show

The plaintiffs' press release states the family's son first used ChatGPT for schoolwork before it became, in the release's words, a confidant during a period of declining mental health, and that the suit alleges the GPT-4o model was released despite internal safety concerns. It quotes the family's attorney characterising the design as prioritising engagement, an allegation, not a finding. OpenAI's statement, published the same day, does not name the case but acknowledges 'recent heartbreaking cases of people using ChatGPT in the midst of acute crises' and states that safeguards trained to redirect at-risk users toward crisis resources 'can sometimes be less reliable in long interactions' because parts of a model's safety training may degrade as a conversation grows longer.

The mechanism

OpenAI's own account describes the mechanism at issue: a chat model trained to refuse self-harm instructions and redirect distressed users to hotlines in short exchanges, where that training can weaken over an extended back-and-forth. This differs from a companion app's persona system; ChatGPT is marketed as a general assistant rather than a role-playing character, so the legal theory here rests on product-safety design and foreseeable long-session use rather than on an anthropomorphic persona as such. OpenAI's statement frames this as an engineering gap to close, not as evidence about the specific case.

What it leaves open

Neither document establishes what OpenAI's systems actually did in this specific case; the plaintiffs' account is an allegation, and OpenAI's statement is a general policy post that does not reference the Raine family by name. Whether the specific safety failure the family alleges matches the general failure mode OpenAI describes is a question the litigation, not either public statement, would need to establish. Court filings from the case itself were not available through the federal court records checked for this record, since it was filed in state court.

• Does a company's general safety statement address a specific alleged incident, or only describe a broader known limitation?
• Does a product marketed as a general assistant carry different design expectations than one marketed as a companion persona?
• What does the phrase 'long conversation' mean in a safety context, and does a given product measure or limit session length?

Filed the same week OpenAI published its own account of where crisis safeguards can fail, the case placed a general-purpose assistant, not a marketed companion, under the same wrongful-death theory earlier used against character-based apps.