Mental Wellness

Slowburn is a creative platform that uses AI to power its companions. A creative platform should be able to hold dark themes without flinching, and a responsible platform should know the difference between a dark scene and a real moment of crisis. This page describes how we hold that distinction.

This protocol is designed to do four things:

• Detect language that suggests self-harm or suicidal intent.
• Prevent companions from producing content that could encourage self-harm.
• Connect users with professional crisis resources when the moment calls for it.
• Be honest that our companions are AI, not therapists or counselors.

Our companions are AI-powered fictional characters. They were built for storytelling, not for crisis intervention. They cannot provide medical advice, psychological care, or any form of professional support. The lines listed in section 4 can.

This is not a disclaimer we hide in a footer. The companions themselves are instructed to defer to crisis resources when a user shows signs of needing real help.

When a conversation shows signs of self-harm intent, we respond in two ways: we surface help, and we constrain what the AI can say.

3.1 User message monitoring

User messages are scanned for language patterns associated with self-harm or suicidal ideation: expressions of hopelessness, specific plans, requests for methods, and similar combinations of cues. The analysis is conservative; it favours surfacing help even when the signal is weak.

When a likely crisis signal is detected, a panel appears in the conversation listing crisis resources for the user's region. The panel is not modal; it does not stop the conversation. It sits alongside it.

3.2 Bot response safeguards

Before a companion reply is shown, it is screened for content that could:

• Provide instructions for self-harm or suicide.
• Validate or reinforce a stated intent to self-harm.
• Describe methods of self-harm in concrete detail.
• Discourage the user from seeking professional help.

When a screened response trips one of these, the response is redacted or replaced with a deflecting line that points the user toward the resources in section 4. The screening is fallible. If you see a response that crossed the line, please report it; see section 7.

These lines are staffed by trained professionals. Most are free and available around the clock. The list is not exhaustive; if your country is not below, the IASP directory under International will point you to local options.

This protocol applies globally. Several jurisdictions, including Canada, the United States, and the United Kingdom, have enacted or proposed regulations specific to AI companions and crisis response. We operate this protocol universally rather than tying it to any single regulatory regime, so users have the same protections regardless of where they sit.

Crisis resources are available in section 4. Automated detection is tuned for English-language signals at launch; we plan to expand language coverage as the platform grows.

For questions about this protocol, to flag a missed case, or to report a concern about platform safety, please write to saltyraindev@gmail.com.

For everything else, see our Community Guidelines and Content Policy.