An Australian Youth Safety Blueprint

Key Findings

• — Pillar 4 asks AI systems used by teens to prevent outputs that encourage emotional overreliance, compulsive engagement, secrecy from parents or trusted adults, or confusion about whether the user is talking to a machine, and to avoid initiating, reinforcing or escalating anthropomorphic behaviours that could lead children to perceive the AI as conscious, emotionally sentient, a romantic partner or a trusted human authority
• — Pillar 5 lists crisis-response commitments: default notification of linked parents when a teen expresses suicidal intent; referral to Triple Zero (000), Lifeline (13 11 14), Kids Helpline (1800 55 1800) and 13YARN; break reminders in long sessions; support for external research; and external advisory councils on mental health and child development
• — Pillar 2 describes age prediction for Australian users, a default safer experience when age cannot be predicted with confidence, and an appeal route through a selfie check by the third-party verifier Persona, which the document says deletes verification data within seven days
• — Pillar 6 lists parental controls the company says all AI systems should offer: account linking from age 13, privacy and memory settings, automatic alerts when a teen's activity suggests intent to self-harm, quiet and study hours, and notice to the parent when a teen changes a parent-set safety setting
• — The document states that nearly nine in ten teens who use ChatGPT use it for learning, information, skill-building or productivity in a given week, and that ChatGPT for Teens began rolling out in Australia in August 2026 as the default experience for users identified as aged 13 to 17
• — Compared with the company's November 2025 US Teen Safety Blueprint (five pillars), this version adds an education pillar and a manipulation-and-dependency pillar; the US text does not use the words dependency or anthropomorphic

Methodology Notes

Company policy position paper, seven pages, dated September 2026 on the cover; the announcement post is dated 18 September 2026. No study method; the single usage figure (nearly nine in ten teens) is asserted without a denominator or survey description. The PDF serves directly from cdn.openai.com and was read in full; the announcement page on openai.com returns 403 to automated fetches and was read from a Wayback raw capture.

Sources