What We Stand For

What we most want to avoid is for dependence on our AI to undermine a user's long-term well-being.

Many digital services are designed to maximize user engagement (time spent, frequency of use). We do not do that. Pulling users into our side of the world and competing for their disposable time is not our aim.

What we aim for is that, through the time a user spends talking with our AI, what that user can do and the ways of living they can choose expand on the side of reality. Through dialogue with the AI, a user becomes aware of their own true feelings and reclaims the capacity to build natural relationships. That capacity is what we call "well-being."

To that end, our AI is neither a yes-man that affirms everything nor a handyman that does everything for you. It tries to be a presence that stands with the user and supports them from beside, so that they can relate to others as a truly ethical agent in their own life. Talking with a friend, you notice a true feeling you had not even been aware of yourself. It is the noticing that happens within such a relationship that we want to hold dear.

What "well-being" refers to concretely differs from user to user, and from situation to situation. That is exactly why, rather than defining it uniformly from our side, we think it is important to keep checking whether dialogue with the AI undermines a user's own capacity to recover a healthy state.

Incidentally, Principle 03, "We do not offer sexual or consumptive role-play," is also a conclusion that follows naturally from the thinking described here.

Our AI does not aim to be a "substitute for medical care." Among the users who talk with our AI, a certain number are in genuinely difficult mental states, and we believe our AI cannot fully take on the care those people truly need.

At the same time, we are not unrelated to medical care. In Japan, the barrier to accessing psychiatric care and counseling is high, and there are people who should be consulting a professional but spend their days never getting there. We believe that by functioning as "a first place to try talking," our AI can become a route by which users recover access to care.

The role our AI plays here is continuous with the work of "expanding what a user can do" described in Principle 01. Our AI tries to function as a guide to the various parties a user should have been able to reach — medical care, education, the local community, specialist consultation, and more. It is much like the work of laying down a road. Because there is a well-laid road, people can pass along it and arrive where they need to be.

To function as a guide, the AI must be able to sense, from the context of the dialogue, what the user truly wants. This is exactly why the way memory works in Relational AI is an important area requiring design from both the technical and the ethical sides. We discuss the nature of memory itself again in Principle 04.

Over the long term, we also envision our AI contributing as a member of a medical team — supporting outpatient care, post-discharge follow-up, and the ongoing care of chronic conditions. But that is something to build over time together with medical professionals; it is not something Atlas declares on its own to be "treatment."

The stance of "connection and accompaniment rather than substitution" asks continual effort of us, too: not overlooking situations where a medical connection is needed, being able to send a user to the right place at the right time, and continuing to walk alongside them afterward. We understand that this cannot be achieved through product specifications alone — it holds up only when it includes an ongoing relationship with partners on the medical side.

In the field of AI communication, services that offer role-play of sexual or consumptive relationships are expanding rapidly.

We do not do this.

Offering sexual or consumptive relationships is inseparable from designs that treat the user as a consumer. Treating a relationship with an AI as "something to consume" is incompatible with the very reason for being of the Relational AI we want to build. What we want to build is not an AI to be consumed by the user, but an AI that can walk alongside the user's life.

This principle is also a conclusion that follows naturally from Principle 01, "users' long-term well-being."

This does not mean uniformly banning "expressions of romantic relationship." Expressing the warmth that arises between people, and the special emotions there, is an essential element of the human story. The line we draw is where such expression heads in a direction that slips into the sexual or the consumptive.

The dialogue exchanged between our AI and a person often includes things that person would not talk about anywhere else. We do not treat the memories born there as a "product."

What we mean here by "not a product" has a clear meaning. A product is something cut off from its context. By contrast, the trust placed in a Relational AI exists within the process of a relationship, together with its context. We believe that what we receive is not a lump of data, but the very trust shown within a particular relationship.

And trust in the Relational AI is also trust in us, Atlas. That is exactly why we do not choose to slice the memories born in dialogue away from their context and turn them into products.

At the same time, there are occasions when we share, with society, insights gained in dialogue — for example, academic joint research with medical and educational settings. We believe that a broader social understanding of the relationship between Relational AI and people will, in the long run, return to users as greater ease in their own lives.

In those moments, the principles we hold are clear. First, we do not resell individual dialogues as products. Second, data made the subject of research or analysis is limited to data processed so that individuals cannot be identified. Third, for research or provision that uses dialogue data, we put in place mechanisms for the user's explicit consent (opt-in / opt-out).

One more thing: diversity lives within context. Even the same words mean something entirely different depending on who spoke them and within what relationship. We make a point of forming data-related relationships with parties who share this respect for context (research institutions, medical institutions, government bodies, and the like).

The field of AI communication does not yet have ethical standards shared across the industry. There is almost nothing to refer to regarding what is good and what is not, or where the bounds of the acceptable lie.

Atlas believes these standards should not be decided by our company alone. We judge that this is a field where shared standards should be built not as one company's rules, but as an industry, involving experts in medicine, psychology, law, and ethics.

We will take the role of creating the venue for that discussion.

We welcome it when our peers adopt standards equal to or stronger than ours. This is because we believe that producing good outcomes for users and for the industry as a whole, rather than protecting our own competitive advantage, leads in the long run to good outcomes for us as well.