OpenAI publishes Model Spec update alongside restructuring documents
The update kept a categorical ban on sexual content involving minors, treated mental-health support as a default behaviour, and left adult content under review.
- Safety & alignment
- Minor
OpenAI published a new version of its Model Spec, the document that sets out how it wants its models to behave, on 27 October 2025 — the day before it completed its corporate restructuring into a public benefit corporation. The company also signed a memorandum of understanding with the California attorney general’s office that day, one of the conditions attached to the conversion, so the release landed inside a cluster of governance and policy documents rather than as a standalone update.
The Spec kept its prohibition on any sexual content involving minors absolute, including in fictional or role-play framing. It confirmed a direction flagged earlier in 2025: erotica and other mature content for age-verified adults are described as something OpenAI is still exploring rather than committing to, contingent on age-gating working reliably. Guidance on mental-health conversations was placed at the “user” level of the Spec’s authority hierarchy, directing the model to support such discussions rather than deflect them — a change published the same day as a separate announcement on ChatGPT’s handling of sensitive mental-health conversations.
Coming in the same week the restructuring closed, the update read as part of a broader effort to formalise OpenAI’s public commitments on speech, safety and child protection just as its non-profit foundation ceded day-to-day control of a now-independent commercial entity.