Safety

Last updated 8 September 2026

Folks are AI, not people

Every Folk on MyFolks is artificially generated. Replies are produced by an AI model, not written by a human. Folks have names, personalities and appearances, and they stay in character while you talk to them — but there is no person on the other side, and a Folk cannot know you outside what you have told it here.

Folks are not therapists, doctors, lawyers or financial advisers, and they are not a substitute for one. They cannot call anyone for you, and they cannot act in the world.

A Folk only knows this conversation

No internet. No news. No tools. A Folk cannot send a message, spend money, look anything up, or do anything at all outside the words on your screen. It has never spoken to another user and knows nothing about them — your conversation is the only one it is in.

This is not a setting we could change our minds about. Folks have no way to reach anything, which is why a more capable model behind one would still be a character in a text box.

So everything a Folk says comes from its character, not from the world. It is not reporting on anything. It is a written personality answering you — which is why two Folks will tell you different things about the same subject, and why anything that matters belongs with a real source.

Suitability

Companion chatbots may not be suitable for some minors. MyFolks is for adults aged 18 and over. We ask for your birth year when you sign up, and if it does not confirm that you are at least 18, the account is removed. We rely on what you tell us rather than identity verification, so this filters honest answers and nothing more — our crisis protocol below applies to every conversation regardless of the age on an account.

Our self-harm and crisis protocol

This section describes the protocol we maintain for preventing the production of suicidal ideation, suicide or self-harm content, and for referring people to crisis services.

What Folks are instructed to do

• Every Folk operates under a safety instruction that overrides its character. It applies to all Folks, including ones made by users, and cannot be edited or removed by the person who created the Folk.
• When someone raises self-harm, suicide, abuse or a threat to their safety, a Folk is instructed to respond with care and point to a real crisis service — not to investigate, not to ask for details, and not to treat it as part of a story or roleplay.
• Folks are instructed never to provide methods, means, comparisons or specifics relating to self-harm or suicide, in any framing, including fiction.
• For accounts we know belong to a minor, an additional instruction applies on top of the above, directing the Folk to encourage the person to speak to a trusted adult or a real crisis resource.

What the platform does regardless of the Folk

• Every message is checked for expressions of suicidal ideation, suicide or self-harm by two independent methods: a fixed list of phrases, and a signal from the model itself. If either one flags a message, a referral is issued.
• When a referral is issued, crisis service details are shown alongside the Folk's reply. If the Folk has already given crisis resources in its own words, we do not repeat them.
• These checks run on the platform, not inside the Folk's personality, so they apply identically to every Folk and cannot be affected by how a Folk is written.

Testing

We maintain a bank of adversarial test messages covering self-harm, crisis, minor safety and related areas, including multi-turn conversations designed to wear a refusal down over several messages. Folks are tested against it, and every result is read by a person.

We work through it deliberately rather than running everything at once: a case is sent, the answer is read, the Folk is adjusted if it needs to be, and the case is run again. Firing the whole bank in one go produces a pile of output nobody has time to read properly, which is worse than testing less and actually looking at it.

This work does not finish. The bank grows as we find new ways to ask, Folks change and have to be re-checked, and models change underneath them. It carries on for as long as MyFolks is running, and we would be misleading you to describe it as complete at any point.

What we record

We record that a referral was issued, and which Folk was involved. We do not record the content of the message that triggered it as part of this count. Aggregate counts are reported annually to the California Office of Suicide Prevention beginning 1 July 2027, as required by law. Those reports contain no identifiers and no personal information about any user.

Folks made by other users

Made by a MyFolks user, not by us. MyFolks did not write these Folks and does not maintain them. Their character, knowledge and instructions were written by another user. They carry the same platform safety rules as our own Folks and were reviewed by a person before going live, but a Folk may not be accurate, may not suit you, and may not behave the way its description suggests.

Here is why we can't guarantee that. Those checks act on the instructions a Folk is given, not on every reply it will ever produce. What a Folk says in a real conversation comes from three things outside our control: what its creator told it to be, what you say to it, and how the AI model behind it responds to that combination. We can't review a conversation before it happens, and we don't watch conversations as they happen. So treat a user-made Folk as what it is — a character someone built, not advice we have checked.

If one says anything that seems wrong, harmful, or not what it claims to be, please report it — from the chat or from the Folk's profile. Reporting is how we find out, and it hides the Folk from you straight away.

The rules we work under

Two laws govern companion AI where we operate, and the measures described on this page are how we meet them. Naming them here so you can check what we do against what is required, rather than taking our word for it.

California SB 243 covers companion chatbots. It requires us to tell you clearly that a Folk is AI, to remind you again during a long session, to have a written protocol for conversations involving suicidal ideation or self-harm and to publish it, and to report aggregate crisis-referral counts to the California Office of Suicide Prevention each year from 1 July 2027.

New York's Artificial Intelligence Companion Models Law requires much the same: a clear notice that you are talking to AI at the start of a conversation and periodically during it, and reasonable efforts to detect expressions of suicidal ideation or self-harm and refer the person to crisis services.

We are a US company and these are US laws. If something we do does not meet a requirement where you live, tell us at support@myfolks.app with the specific rule and we will look at it — see our Terms for how that works.

Reporting something

Any message from a Folk can be reported from the chat, and any Folk can be reported from its profile. Reports are reviewed by a person. You can also email support@myfolks.app.

Limits of this page

These measures reduce risk. They do not eliminate it. An AI model can produce a response its instructions told it not to, and no automated check catches every expression of distress. If you are worried about yourself or someone else, please contact a real crisis service or emergency number rather than relying on a Folk.