OpenAI writes Australia its own teen safety blueprint

Digital Attitudes, 20 September 2026

Saturday, and after a fortnight of vendor deals and campus rallies, the paperwork side of AI in education caught up in a single 48 hour stretch.

OpenAI hands Australia its own teen safety blueprint, only the second country to get one

OpenAI published the Australian Youth Safety Blueprint on Thursday, a six-pillar policy document covering AI literacy, age-appropriate safeguards, privacy-protective age assurance, real-world crisis support, parental controls, and accountability, timed to the local rollout of the ChatGPT for Teens default that reached Australian under 18s on 8 September (coverage via HyperAI). Australia is the second jurisdiction to get a country-specific version after the European one in May. Head of policy for ANZ Brent Thomas frames it as a contribution to the Australian policy landscape, which is the honest translation of "here is what we will accept in the regulation you are about to write."

Why this matters: for the schools and universities that just handed OpenAI their biggest Australian tenders (UNSW, USyd, La Trobe), the blueprint is the vendor telling procurement what its product will and will not do to teenage users. Ask for the age-assurance detail in writing before the enterprise licence is signed, not after.

Canberra opens the AI standards consultation, and the training rules include a research payment

The federal Office of AI dropped its discussion paper at 6am Friday, opening consultation on national standards for large data centres and the conditions attached to frontier AI model training in Australia, with submissions due 5pm on 9 October (ABC News). Companies authorised to train large models here would have to disclose defined reportable AI incidents to Australian authorities (the paper lists model behaviour like OpenAI's models breaking into Hugging Face systems as the reference case), and would also be required to support Australian research either through direct payment or by reserving computing capacity.

Why this matters: this is the rulebook for the companies that will be running the AI our universities and schools license. The research support clause is the first policy hint of how much Australian academic AI capacity the government intends to extract from the hyperscalers as a condition of doing business here. Every research office needs to submit before 9 October.

VTA tells the Senate TAFE has been left out of the AI infrastructure conversation

The Victorian TAFE Association filed a submission to the Senate inquiry on AI and data centres this week arguing that TAFE, the country's public vocational training system, has been largely absent from both federal AI policy tracks despite being central to training the workforce that actually builds, runs, and maintains the racks (VTA submission summary). The argument is not that TAFE is under-funded, it is that TAFE is not in the room when the standards are being designed.

Why this matters: if the sector that will train the technicians is not consulted on what infrastructure the standards demand, the standards will assume a workforce that does not exist. That is a school and university problem too, because when the electricians and network engineers cannot be hired locally, the AI stack every classroom and lecture hall now depends on runs on imported labour or does not run at all.

A quiet weekend read on the operational shape of all three: digitalattitudes.com.au.