Meta's Muse Wonder remembers you across chats: schools and employers should ask what it keeps

On 12 October 2026 Meta released Muse Wonder and Muse Wonder Full, which it describes as its most capable models, trained at its Prometheus data centre. Muse Wonder is available now through the Meta Model API and Muse Code. Muse Wonder Full is available to Muse Code subscribers. Meta says Muse Wonder will soon power experiences in its Muse assistant, rolling out first to subscribers. TechCrunch reports API pricing of $4 per million input tokens and $20 per million output tokens, and calls it Meta's first frontier model.

Meta's headline feature is "Active Memory", which it describes as "carrying important context about you across conversations and actively retrieving the details relevant to the task at hand". A second feature, Active Focus, has the model keep working towards a goal, try other approaches when blocked, and check in before important actions. Meta calls Muse Wonder "our most aligned model to date" and says it was trained for "careful handling of private information". A full safety report is promised "soon".

Most people have only used AI that starts every conversation from nothing. An assistant that carries context about you from one conversation to the next is a different kind of tool. It can know what a teacher struggles with, what a manager is worried about, or what a family has asked for help with. That makes it more useful, and it makes the question of what it keeps matter far more. Meta's model card says the models are "not intended to be used by individuals under the age of 18" and leaves deployers to assess the risks of such use. A line in a model card is not a control that a school or a parent can see. The card also says that, without mitigations, Muse Wonder Full may reach Meta's high-risk threshold for chemical and biological risk; with safeguards, Meta rates residual risk as moderate or lower.

Dan Fitzpatrick's reading is that memory is the feature to watch, more than the benchmark scores. Memory is what turns a chatbot into an assistant people lean on, and it is also where more of the thinking can quietly move into a system nobody can inspect. The announcement describes what the model remembers. It does not set out what users can see, correct or delete, and those are the questions that decide whether trust is earned. None of this is a reason to refuse the technology. Paralysis is a bigger risk than adoption, and teachers and workers are already ahead of policy. It is a reason to adopt with the right questions in hand.

For school leaders and business leaders, the practical steps are these. Before anyone uses a memory-enabled assistant for work, find out whether memory can be switched off, what is stored, who can see it, and how it is deleted. Add persistent memory to your acceptable-use and data protection policies now, rather than after the first incident. Ask vendors to answer in writing. Teach pupils and staff one habit: ask what a tool knows about you before you tell it more. Parents should check the settings of any assistant their children use.

Dan Fitzpatrick works with school leaders and employers on how they adopt AI day to day, and writes and speaks on what it means for learning, work and trust.

Dan Fitzpatrick is available for interview and comment on this story. He is available today for live or pre-recorded radio, TV, podcast and print interviews, and has a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io