Product news: GPT-5.6 Luna expands free text access while adding explicit under-18 safeguards

OpenAI's August 6 product announcement expands ordinary access to GPT-5.6 while making several youth-safety boundaries visible. GPT-5.6 Luna was scheduled to become the default model for Free and Go users during the week of the announcement. Starting the following week, those users were to receive unlimited text chats and a Think button for questions that need more reasoning time, subject to abuse guardrails. File uploads, image generation and other tools would continue to have limits. Plus and Pro users could use an updated GPT-5.6 Sol and a slider for choosing how much thought the model applies from August 6.

For education, the access change matters because a general free account can be easier for learners to reach than an institutionally managed system. Unlimited text does not mean unlimited capability, guaranteed availability or school approval. It also does not turn a consumer account into ChatGPT for Teachers or ChatGPT Edu. Those products have different verification, administration and data-governance arrangements. A school should therefore distinguish what students can access independently from what it authorizes for coursework, protected information or assessment.

The announcement also details training and system protections for users OpenAI believes are under 18. OpenAI says the model is trained to avoid romantic roleplay, age-restricted challenges and presenting itself as a substitute for real-world relationships. It describes age-appropriate boundaries around sexual content, eating disorders and body-image risks, age-restricted goods, dangerous activities and graphic violence. When a teenager may need support, the model is intended to encourage connection with trusted people. These are vendor-described safeguards, not evidence that every risky interaction will be detected or resolved well.

Schools should not treat a model's age-aware behaviour as a replacement for safeguarding practice. Age inference can be imperfect, students may share accounts or omit context, and a fluent refusal can still leave a distressed learner without adequate help. A deployment plan should identify approved accounts, escalation routes, human contacts, reporting procedures and when AI should not be used. Teachers also need clarity about whether chat records, files or images enter a school-managed workspace or a personal account.

A useful pilot can separate access, safety and learning outcomes. Students might use text chat for low-risk brainstorming or explanations, then verify claims against course sources and complete an independent task. Schools can test age-inappropriate prompt categories with authorized synthetic scenarios, record false negatives and false positives, and confirm that support messages lead to real human resources. They should also monitor whether unequal access persists through device, language, tool and file limits.

The release lowers one access barrier and states stronger youth boundaries, but neither change proves educational benefit. For Hong Kong educators, the practical question is whether a clearly governed workflow helps learners reason, verify and seek human support while keeping consumer access separate from institutional responsibility.

Any local trial should publish its account assumptions, age controls, escalation route and tool limits so families can distinguish tested safeguards from marketing language. That record should be refreshed whenever the default model, rate limits or safety behaviour changes.