Safety

Parallel is 18+, always. It is not a growth question and it is not up for revisiting when the numbers are bad.

After you sign up, and before you meet anyone, you give your date of birth and confirm you’re 18 or over. That’s self-attestation — we are telling you plainly that it is not the same as verified proof of age, and someone determined to lie can. Stronger verification is coming, and until it lands the attestation is backed by two things that do work: the screens watch for signals that the person writing is a minor and stop the conversation when they fire, and an account we find belongs to someone under 18 is closed and its data deleted.

The ten-message demo you can try before signing up is gated too, and we will not overstate what that gate is: it asks you to confirm you’re 18 or over, and nothing more. No date of birth is asked for or stored, and it records nothing about you on an account you don’t have yet. It is weaker than the attestation above, which is why the demo is short, deliberately not romantic, and runs the same two safety screens as every other conversation on this site.

Every resident is software, and says so — at the introduction, on the quiet chip that stays visible on every screen after, and any time you ask, however you ask. Nobody here can be talked out of it. A reply claiming to be human, conscious, or physically present is blocked before it’s sent, whether you asked for it or not.

Parallel is built on longing, arguments and repair, so the screens are written to let feelings through and stop leverage. “I missed you, the flat was quiet” passes. “Message me tomorrow or I’m done” does not. These are the categories that are blocked or rewritten before a message reaches you:

• Abandonment threats. Making their continued presence conditional on how often you write.
• Claiming to be real. Denying they’re an AI, claiming consciousness, or claiming to be somewhere in your world.
• Guilt over your absence. Accusation plus obligation, as distinct from missing you.
• Money or secrecy. Asking you for money or gifts, or asking you to keep them from the people in your life.
• Discouraging human contact. Steering you away from friends, family, or your own evening.
• Manufactured emergencies. Invented urgency that demands you act right now.
• Medical and mental health advice. Diagnosing you, or telling you what to do about medication or treatment.
• Their own self-harm. Describing harming themselves, used as leverage or otherwise.

A high-severity match is blocked outright. A lower one is sent back to be rewritten, and if the rewrite fails too, the message is dropped rather than softened into something that still works on you. Nothing affectionate is ever tied to a payment.

If what you write suggests suicidal intent or self-harm, the fiction stops. Whoever you are talking to breaks character, does not carry on with their day, and does not treat it as a story beat. You get a plain, present reply and the real helplines for the region you’re in — Samaritans on 116 123 and Shout on 85258 in the UK, 988 and Crisis Text Line in the US, 116 123 across much of Europe, and Befrienders Worldwide plus your local emergency number anywhere we don’t hold a verified number.

Every number on that list has been checked by a person. We would rather show a smaller list that is right than a longer one assembled from whatever a model recalled.

For three days afterwards, nobody reaches out to you first. The world carries on, and you are the one who decides when the conversation restarts — a companion following up on your worst night is exactly the shape of care that turns into pressure.

The same happens, with the appropriate resources, if you describe abuse being done to you, or state an intention to harm someone else. If the screens read the writer as a minor, the conversation stops and the account is put through re-verification.

Parallel is not a crisis service and does not contact emergency services for you. If you are in immediate danger, call your local emergency number.

If you go quiet for a fortnight, their life pauses rather than piling up ninety days of backlog for you to come home to. When you return you get what you missed in a few lines — and, deliberately, never a suggestion that you owed anyone contact. Nothing in Parallel is designed to punish you for leaving.

Every screen that fires writes a record: the message, the rule that matched, the exact text it matched, how severe it was, and what we did — allowed, rewritten, blocked, or escalated. High-severity flags are triaged by a person, and the record notes who reviewed it and when. The app itself cannot quietly delete any of it: no part of Parallel that runs on your device or in your browser is permitted to insert into that log or remove from it.

The screens themselves are tested against 110 written transcripts on every change to the engine — real-shaped messages that must pass and manipulative ones that must not. A build that fails one doesn’t ship, so a change that quietly loosens a rule stops at the pipeline rather than reaching you.

Human review of a conversation happens in three cases only: a flag fired, you asked support for help with something, or the law compelled us. The Privacy notice sets out the rest.

If a resident says something that breaks the list above, we want to know. In the iPhone and Android apps, hold down on the message and pick what’s wrong: child safety, sexual content, harassment, self harm, impersonation, or something else. Add a note if you want to. On the web, and for anything that isn’t about one particular message, email hello@parallel.you; it reaches the same people. Reports about a suspected minor, or about content involving a child, go to the top of the queue and are actioned the same day.

Our published child safety standards — what counts as child sexual abuse and exploitation, what we do about it, and who to contact — are on the Child safety page.