Safe to explore.
Private to grow.

We put children’s safety, healthy development, and rights at the center of every product decision. When a feature does not yet meet our safety bar, we adjust it, limit it, or delay it. Trust is built before anything is scaled.

The Explorer profile is designed to determine an appropriate age group from the start — collecting no more precision than necessary — and experiences are designed by age group: in content, language, how the AI interacts, and how adults are involved. An experience that suits an older child may need to be designed differently for a younger one.

BlueStar is an AI companion, and children are told so in language that fits their age. BlueStar never pretends to be human.

AI can help spark questions, explain, create, and experiment. AI can also misunderstand, give inaccurate information, or miss something important. So children are encouraged to question the answers — and to turn to adults when they need to.

BlueStar is designed to support relationships in real life, and never to do three things: ask a child to keep worrying interactions secret; claim to be the only one who understands them; or make them feel guilty for leaving.

BlueStar is available whenever a child needs it, but it is not designed to be ever-present. Available, not omnipresent. A child always has the right to end a conversation, change direction, and give their time to their own life.

OZOverse does not use fear of missing out, social comparison, shame, or a child’s affection for BlueStar to keep them on a screen or to drive purchases. Time spent is not a metric we optimize.

We build experiences that help children think for themselves, try their own way, and gradually act more independently. Personalized suggestions exist to widen what a child can explore: they do not box a child into old interests, impose one model of success, or draw fixed conclusions about what a child is capable of.

We respect differences in culture, circumstance, ability, and self-expression. Children are treated with dignity, even when the way they learn or the choices they make differ from the majority.

Technology should expand a child’s life, not shrink it into a screen.

Our approach combines selecting appropriate content and tools, limiting what the AI can do, testing responses, and assessing potentially harmful situations in advance. Content involving violence, exploitation, age-inappropriate sexual material, self-harm, or dangerous behavior has dedicated protections. Sensitive questions are handled calmly, respectfully, and at a level the child can understand.

When an experience leads into the real world, suggestions take into account age, materials, and setting. Activities involving water, electricity, heat, heights, sharp tools, or chemicals always require an adult’s judgment and supervision — and the product says so, clearly, to the child.

There is one situation we do not look away from: when a child shares signs that they are in serious danger — abuse, violence at home, or thoughts of hurting themselves.

Our principle in these situations: respond calmly and age-appropriately, and guide the child toward the right people. The path to help is determined by each child’s circumstances — because the safest help does not look the same in every family. In cases of serious danger, OZOverse may need to limit privacy in order to protect the child, under a clear safeguarding process. Children are told, in words that fit their age, that things like this are not secrets between them and an AI — because they deserve to be protected by real people.

The mechanisms for recognizing these signals and assessing their severity and context are being built and tested together with the appropriate expertise.

We collect only the data needed for purposes we specifically explain, and we do not keep a child’s personal data longer than necessary for those purposes. We do not sell children’s personal data, and we do not advertise to children — in any form.

Account information, conversations, voice, images, and creations can reveal private things about a child. Every category of data is weighed for its purpose, who can access it, and how long it is kept — before it is ever collected. A child’s profile and journey are designed to be private by default; sharing is always an active choice, with the appropriate permission.

We are building the mechanisms for families to know clearly when a microphone or camera is in use, whether data is stored or sent to service providers, and how to turn those functions off.

The security of children’s data is part of the design, not a patch on top. The system is being built on these principles: encrypting data in transit and at rest; limiting internal access by role and task; logging access; and testing the system regularly.

If a security incident affects user data, we will notify affected families promptly — explaining what happened, what data was involved, and what we are doing about it — meeting, and where possible ahead of, what the law requires.

For young children, the involvement of a parent or guardian is an essential part of how OZOverse is designed. Before any data processing that requires consent, we explain clearly what data is collected, what it is used for, and who it is shared with.

We are building a parental consent verification mechanism that meets the requirements that apply in the markets where we operate (such as COPPA in the United States), along with tools for families to review information, request corrections or deletion, withdraw consent, and stop further collection — with any effects on features clearly explained.

Access for parents, teachers, and other supporting adults is specifically defined. And children are told, in age-appropriate terms, what adults can see — because protecting a child must go hand in hand with respecting their growing voice and their own space. One important limit on that space is the serious safety situations described in section 04.

The Living Atlas keeps meaningful experiences, questions, and creations so that a person can return to them and keep exploring. Remembering has limits by design: a passing conversation does not automatically become a lasting memory.

OZOverse does not use a child’s history to permanently label them, rank their worth, or diagnose them.

We are building the mechanisms for every AI impression to be reviewed and corrected, and a deletion process that explains clearly how data in use, backups, and any legally required retention exceptions are handled.

OZOverse connects multiple AI models and tools. Before any capability enters a child’s journey, it is evaluated for its value, appropriateness, safety, and how it handles data. We pass along only the information each task requires, and every tool has a clearly defined scope; actions such as sharing externally, publishing content, or making payments always require the appropriate permission.

We do not use children’s personal data to train foundation AI models — ours or any provider’s. Consent to use a feature is never treated as consent to use data for AI training or for any other purpose; every other purpose, if any, is explained separately and requires separate consent. Our Privacy Policy defines and clearly separates four different processing activities — AI model training, personalization, safety evaluation, and product improvement — so that “no training” is never misread to mean that data is never processed to personalize a child’s experience or to keep that child safe.

OZOverse’s responsibility to protect its users extends to how we select, bind, and oversee our providers.

AI changes, and so does the way people interact with it. Safety evaluation therefore happens before release and continues through operation: testing hard scenarios, assessing bias risks, controlling data access, and running an incident process. Every change of model or tool is reviewed for its impact on children’s experience.

We measure OZOverse’s design against outside standards — the children’s online protection regulations of the markets where we operate, and age-appropriate design codes.

We are building a mechanism to bring independent expertise in child development, education, and safety into how the product is evaluated, rather than relying on internal review alone. We are also building a reporting channel for families to flag concerning content, inappropriate behavior, or privacy issues — with a process that states clearly how reports are handled and when to expect a response.

We communicate known limitations honestly, announce significant changes, and fix what falls short.