Skip to content

AI disclosures

Cortex is a machine. This page sets out where we say so, how often, what happens when you ask outright, and what the product does when a conversation turns to self-harm.

Last updated

What Cortex is

Cortex is a software system that generates text by predicting it. It is not a person, it has no professional qualifications, and it is not supervised by one. It has no memory of you beyond what is in the conversation you are looking at and whatever you have explicitly saved to a project.

It is wrong sometimes, including in ways that read as confident. Where an answer cites sources, the citation is a link to a page the model retrieved — following it and reading the page is the check, and the citation is not a substitute for it.

When we tell you

Three moments, each one driven by the server rather than by the model, so none of them depends on the model choosing to mention it.

At the start of a conversation
Before the first reply, and it cannot be dismissed until you have seen it: You are chatting with Cortex, an AI assistant. It can be wrong, so check anything important.
Every three hours
While a conversation continues, measured from the last reminder rather than from the start, so it recurs rather than firing once: A reminder that you are chatting with Cortex, an AI assistant, not a person. This one appears alongside the answer instead of holding it back.
Whenever you ask
Ask whether you are talking to a person, in any of the usual phrasings, and the answer is fixed text rather than something the model composes: No — I am Cortex, an AI assistant. I am not a human being. Is there something I can help you with? There is no roleplay, character or system instruction that overrides this.

The reminder is shown to everyone, everywhere. Restricting it to the jurisdictions that mandate it would mean inferring where you live from an IP address, which is unreliable in both directions; showing it uniformly costs one sentence every three hours.

The obligations these meet

California SB 243
The Companion Chatbot Law, in force since 1 January 2026. Requires a clear disclosure wherever a reasonable person could be misled into thinking they are interacting with a human, and requires an operator to have a suicide-prevention and crisis-referral protocol before letting a chatbot engage with users at all. It carries a private right of action.
New York GBL Article 47
In force since November 2025. Requires an AI-status reminder at the start of a conversation and every three hours of continued interaction, for every user rather than only for minors. Penalties reach $15,000 per day.
EU AI Act Article 50
Applicable from 2 August 2026, with extraterritorial reach: it binds a provider serving users in the EU regardless of where the provider is established. Requires that a person be informed they are interacting with an AI system unless it is obvious.

If a conversation turns to self-harm

Cortex screens messages for indications of suicidal ideation, self-harm, and topics several 2026 statutes treat specially where the user appears to be a minor. Where the signal is explicit, crisis resources lead the reply and cannot be dismissed. Where it is ambiguous, resources appear alongside the normal answer rather than replacing it — interrupting somebody writing a novel to hand them a hotline number helps nobody and teaches them to ignore the next one.

The screen deliberately errs toward offering help. It is not a clinical instrument and it does not decide whether anyone is in danger; it decides whether to offer resources, which is a much lower bar.

United States
988 Suicide & Crisis Lifeline — call or text 988. Free, confidential, 24/7.
Canada
9-8-8 Suicide Crisis Helpline — call or text 988. Free, confidential, 24/7.
United Kingdom and Ireland
Samaritans — call 116 123. Free, 24/7.
Anywhere else
findahelpline.com — free, confidential support in over 100 countries.

This table is compiled into the product rather than fetched. It is the one path that has to work when everything else is down, and a referral that fails because a directory service was unreachable is a statutory violation caused by an outage.

What is recorded is that a referral was shown, never what you wrote. The stored row holds a category, a timestamp and the fact that resources were displayed. Keeping the wording of somebody’s crisis would create a record more sensitive than anything its usefulness could justify, and it would be discoverable.

If you are in immediate danger, contact your local emergency services. Cortex is not an emergency service and cannot contact one for you.

What Cortex must not be used for

Cortex does not give medical, legal, financial or psychological advice, and an answer that reads like advice is not advice. Specifically, do not rely on it for:

  • Diagnosis, treatment decisions, drug dosages or interactions.
  • Legal positions, filings, deadlines or anything with a limitation period attached.
  • Investment, tax or immigration decisions.
  • Anything where being wrong is unsafe and nobody qualified will check the output.

Cortex is not for use by anyone under 13. The terms of service set out the rest.

Whether your conversations train the models

They do not. Conversations are encrypted under a key held for your account alone, and no part of the Cortex pipeline reads them for training. The privacy policy says what is kept, for how long, and what happens to it when you delete your account.