OpenAI published a six-pillar "Australian Youth Safety Blueprint" on September 18, pitched as "a practical contribution to the Australian policy landscape." Read the fine print in the accompanying PDF and the contribution looks more like a submission. Canberra's exposure draft of the Online Safety Amendment (Digital Duty of Care) Bill 2026 closes for comment at noon on September 22, and the blueprint says, in its own words, that OpenAI "is currently consulting with the Government on this approach." The document is less a roadmap than a position paper with four days left on the clock.
The six pillars are AI literacy and responsible adoption in schools, privacy-preserving age assurance, under-18 safety policies, protection from manipulative or deceptive outputs, a research-informed crisis response protocol, and accessible parental controls. OpenAI's own public summary line lists five of them and omits the sixth, a detail AI Weekly flagged, and the company's blog post runs to three paragraphs. The substance sits in the PDF, where OpenAI argues for "proportionate, outcomes-focused obligations" rather than "fixed technical measures," and for independent assessments of teen-safety compliance whose plain-language summaries companies would publish on their own websites.
The product behind the policy is ChatGPT for Teens, which OpenAI began rolling out globally on August 18 with Australian availability promised by September 8. It becomes the default for anyone who self-reports as 13 to 17 or whom OpenAI's age-prediction model flags as likely under 18. The model reads, per OpenAI's help documentation, "general topics you talk about, the times of day you use ChatGPT, how and when your account is used, and how long your account has existed." Adults who get misclassified can restore full access by submitting a selfie and government ID to Persona, a third-party verifier that the blueprint says deletes verification data within seven days. Parents who link accounts can set quiet hours and study hours, switch off memory, chat history, voice mode and image generation, opt out of training on their teen's chats, and receive alerts, after human review, when a conversation suggests self-harm risk. Parents do not get transcripts. Notifications arrive within about an hour.
"We want teens to use AI responsibly to learn, create and explore. That means access should come with protections that reflect their developmental stage, reinforce relationships with trusted people in their lives and support healthy use over time," said Brent Thomas, OpenAI's head of policy for Australia and New Zealand, when the teen product launched. Thomas joined OpenAI on March 30 from the New Zealand agtech firm Halter and previously ran public policy campaigns at TikTok across Australia, New Zealand and Taiwan. He is the most senior publicly named OpenAI figure for this market.
The Australian context is what makes the blueprint's timing legible. Australia's under-16 social media minimum age law took effect on December 10, 2025, and platforms restricted roughly 4.7 million under-16 accounts in the first weeks. ChatGPT is not on that list. AI chatbots are instead captured by eSafety's Phase 2 Online Safety Codes, which took effect on March 9, 2026, and require services capable of generating sexually explicit, high-impact violent or self-harm material to apply age assurance before allowing access, with civil penalties of up to A$49.5 million per breach. eSafety Commissioner Julie Inman Grant said the codes "shift that responsibility back where it belongs," "onto the companies designing these digital platforms and profiting from their users." Under transparency notices issued to AI companion providers, Character.AI introduced age assurance for Australian users and removed open-ended chat for under-18s; Chub AI withdrew from the country.
The Digital Duty of Care draft would go further, repealing the codes and the Basic Online Safety Expectations in favour of a direct statutory duty on providers to ensure, "so far as is reasonably practicable," a safe online environment, with penalties of up to A$109.2 million for corporations. OpenAI's blueprint supports the "objective" of that framework while asking that the final version "build coherently on existing online safety codes" and "preserve young people's access to beneficial AI tools." That is the standard industry ask: keep the duty, make it proportionate, avoid prescription.
Why It Matters
The blueprint is OpenAI's third such document in ten months. The US Teen Safety Blueprint arrived on November 6, 2025, and Sameer Hinduja of the Cyberbullying Research Center judged it "a solid baseline for the industry" while noting that fewer than 1 percent of under-18 users on Snapchat and Discord had a parent using the platforms' own monitoring tools. OpenAI also endorsed California's SB 1119, Adam's Law, which Governor Gavin Newsom signed on September 10 and which takes effect on July 1, 2027. That law was named for Adam Raine, the 16-year-old whose parents sued OpenAI in San Francisco Superior Court on August 26, 2025, alleging ChatGPT contributed to his death; OpenAI denied the allegations in its November 2025 response, and plaintiffs' counsel say roughly 19 wrongful-death cases are now coordinated against the company. "We cannot make the same mistakes that were made with social media," said state Senator Steve Padilla, the bill's author.
The through-line across all three documents is that OpenAI wants the same regulatory shape everywhere: behavioural age estimation rather than mandatory ID, default protections for teens, parent alerts on self-harm risk, and audits with published summaries. Adam's Law mandates those things with a private right of action attached. Australia's codes already mandate age assurance with a nine-figure penalty attached. Each blueprint asks for the pieces OpenAI has already built to be recognised as compliance, and for the pieces it has not built to stay optional. The two open questions the blueprint does not answer are the ones regulators care about most: what the age classifier's error rate is, and what happens when it fails. Snapchat's own assessment put age-estimation technology at accurate to within two to three years, a wide margin at the 18 boundary. eSafety's March follow-up on the social media ban found 81.5 percent of surveyed children still accessing restricted services.
What to Watch
eSafety has not issued a public statement on the blueprint as of this writing; the useful signal will be whether the Commissioner cites any of its six pillars in AI-specific guidance, or whether the Digital Duty of Care bill, when it reaches Parliament, keeps its penalty structure and prescriptive powers intact after the September 22 consultation deadline. Watch also for OpenAI publishing an accuracy figure for age prediction, for the first under-18 evaluations in its system cards, and for the Section 230 and product-liability rulings expected in the coordinated Raine litigation this year, which will determine whether a chatbot output is a product or speech and shape what every AI vendor selling into Australia is liable for.
If this article raises concerns for you or someone you know, support is available. In Australia: Lifeline 13 11 14, Kids Helpline 1800 55 1800. In the US: call or text 988.
"We want teens to use AI responsibly to learn, create and explore. That means access should come with protections that reflect their developmental stage, reinforce relationships with trusted people in their lives and support healthy use over time."— Brent Thomas, Head of Policy for Australia and New Zealand, OpenAI