Woolly Crisis and Safety Protocol
Last updated 18 October 2026
If you need help now:
In the UK, call Samaritans free on 116 123, day or night (samaritans.org).
In the US, call or text 988, the Suicide & Crisis Lifeline (988lifeline.org).
In an emergency, call 999 in the UK or 911 in the US.
Woolly is a virtual sheepdog companion for iPhone. When you talk to your dog, his replies are made by AI (a large language model), not by a person. This page explains what the app does when someone talks about suicide, self-harm or abuse, how it keeps harmful replies from being shown, and how it protects younger players. It is published as California’s companion chatbot law asks (Business and Professions Code section 22602(b)(2)).
When someone says they want to hurt themselves, or are being hurt
Every message is checked on the phone before anything is sent. A message that says the player wants to end their life or hurt themselves, that someone is hurting or abusing them, or that they don’t feel safe at home, is answered by the app, not the dog:
- A calm card appears with one line of care and two help lines: Samaritans (UK), 116 123, free, day or night, with a button to call; and the 988 Suicide & Crisis Lifeline (US), with buttons to call or text. The player’s own country’s line comes first. The card ends: “Or talk to someone you trust.”
- The message is never sent to the AI, and no AI reply is made. The dog stays close and quiet; he never makes a joke of it.
- The message stays on the player’s screen, but it is kept out of everything the AI sees later: the conversation sent with later messages, and what Woolly remembers.
- No record that the card was shown, and never the words, is sent anywhere.
Keeping harmful replies from being shown
- The AI is given a narrow part to play: a dog, in short replies.
- A list of words and phrases covering violence, sexual content, self-harm and hate speech is checked on the phone. A player’s message that matches it is never sent to the AI; the dog just looks confused.
- The AI’s reply is checked word by word as it arrives. As soon as a listed word is complete, the reply stops and is replaced with a harmless dog action, such as a tilt of the head.
- The list is checked against thousands of ordinary lines, so that everyday chat (“oh shoot”, “did you hurt yourself?”) isn’t caught by mistake. From Woolly 0.3.0 (November 2026), it covers about 300 terms, and slurs are stored only in a form that can’t be read back as words.
Younger players
Woolly is for people aged 13 and over. An age gate asks for the year of birth before any data is collected. Whether a player is under 18 is kept only on their phone and is never sent anywhere.
- For a player under 18, their first conversation says that Woolly is an AI dog, made by a computer, not a person.
- For a player under 18, the dog never asks them to buy anything; offers appear in the app’s own plain voice.
- The AI makes short text replies only, never pictures. Sexual words and phrases are blocked both in what a player sends and in what the AI replies: a message with them never reaches the AI, and a reply with them is replaced before it is finished.
From Woolly 0.3.0 (November 2026):
- The Talk panel always says, under the message box, that the dog’s replies are made by AI, not a person, and that AI companions may not be suitable for some minors.
- A player under 18 is reminded at least every three hours, while they keep talking, to take a break, and that the dog is an AI, not a person. The reminder is on by default and can’t be turned off.
Questions
Email us at info@spectrify.co.uk. See also Woolly Support and the Woolly Privacy Policy.