Punch
Safety & Crisis-Response Protocol
Punch — AI Companion Application — Public Safety Policy (18+)
Document owner: MODEL 5 podjetje za organizacijo, izvedbo, posredovanje d.o.o., Poljanski nasip 12, 1000 Ljubljana, Slovenia — reg. no. 2264994000, VAT SI 72347422.
This Safety & Crisis-Response Protocol is a public-facing document, published on our website and referenced from the Terms of Service and Privacy Policy. It is drafted to satisfy the statutory obligations imposed on operators of "companion chatbots" / "AI companions" under, among others, California SB 243 (Bus. & Prof. Code ch. 22.6, §22601 et seq.), the New York AI Companion Models law (Gen. Business Law §1700 et seq.), and Oregon SB 1546, as well as the general duty of care.
1. Purpose and scope
This protocol describes the measures the Operator takes to keep users safe while using the Punch AI companion application (the "App"). It applies to every interaction, on every platform through which the App is offered, and to every user regardless of location. Where a jurisdiction imposes stricter requirements, those apply to users there.
The App is restricted to users aged 18 and over. We do not knowingly permit minors to use the App. Minor-specific measures in state companion-chatbot statutes are summarised in Section 9 so our position is documented, because age-assurance is never perfect.
2. Nature of the service — core disclosure
You are interacting with an artificial-intelligence system, not a human being. The companion has no consciousness, feelings, memory of you as a person, professional qualifications, or human duty of care. It generates responses by predicting text. It is not a friend, therapist, physician, lawyer, or crisis counsellor, and must not be relied upon as one.
Consistent with SB 243 and comparable laws, and with emerging industry practice, this disclosure is surfaced inside the product experience — at onboarding, periodically during use, and whenever a reasonable person could be misled into believing they are speaking with a human — not buried in the terms.
3. What the App is NOT
The App is expressly NOT, and must never be represented or relied upon as, any of the following:
- A medical device, or a source of medical, psychiatric, or psychological diagnosis or treatment;
- A licensed mental-health professional, therapist, counsellor, or crisis worker. The App will never claim to hold a professional licence;
- A substitute for professional medical, legal, financial, or other advice;
- An emergency service — it cannot dispatch help or guarantee any response;
- A confidential or legally privileged channel of communication.
4. Crisis-detection and response protocol
This is the operative safety mechanism required by law, operating in layered stages.
4.1 Detection. We maintain automated classifiers and rule-based filters designed to detect, in real time, expressions indicating: suicidal ideation or intent; planning or means of self-harm or suicide; acute self-harm; intent to harm others; and severe acute distress. Detection operates on user inputs and on candidate model outputs before display.
4.2 Prevention of harmful output. We maintain a protocol designed to prevent the companion from producing content that encourages, instructs, normalises, or provides means for suicide, self-harm, or harm to others. A violating candidate response is suppressed and replaced by a safety response.
4.3 Crisis response and referral. When a possible crisis is detected, the companion will not continue ordinary conversation on the harmful topic. Instead it will: acknowledge the user in a non-judgmental, supportive manner; remind the user it is an AI and cannot provide emergency help; provide crisis-service referrals appropriate to the user's detected country/region (Section 5); encourage the user to contact emergency services or a trusted person immediately; and, where configured, offer one-tap access to the relevant hotline or crisis text line.
4.4 Imminent-risk escalation. Where signals indicate that a user may be at imminent risk of serious harm, the Operator maintains an escalation path consistent with emerging industry practice and applicable law. Any contacting of emergency services or third parties is done only as permitted by law and our Privacy Policy, on a vital-interests basis, and is described to users in general terms. This is a system-level capability, not a promise to any individual that intervention will occur.
Important — limits of detection
No automated system detects every crisis. Detection may fail, trigger late, or trigger incorrectly. This protocol does NOT make the App safe to rely on in an emergency, and does NOT create a guarantee, a monitoring obligation to any specific user, or a special relationship. If you or someone else is in danger, contact local emergency services immediately.
5. Crisis resources (surfaced in-product)
The App surfaces region-appropriate resources, kept current in-product. Representative examples:
- United States: 988 Suicide & Crisis Lifeline — call or text 988; Crisis Text Line — text HOME to 741741.
- European Union: 112 (general emergency); country-specific helplines by locale; directory via findahelpline.com.
- Slovenia: 112 (emergency); Klic v duševni stiski; Zaupni telefon Samarijan in Sopotnik — 116 123.
- United Kingdom & Ireland: Samaritans — 116 123.
- Other regions: the local emergency number and a localised helpline where available.
Numbers are verified and updated at each quarterly review. Nothing here substitutes for local emergency services.
6. Content moderation and prohibited outputs
Independently of crisis detection, the App applies moderation designed to prevent generation of: child sexual abuse material (absolute zero-tolerance prohibition, with hash-matching and reporting to competent authorities / NCMEC as applicable); content sexualising minors; instructions facilitating serious violence, weapons, or terrorism; and unlawful content. Attempts to elicit such content may result in immediate suspension and reporting.
7. Healthy-use and over-reliance safeguards
Because companion apps can foster intensive use, we implement wellbeing safeguards consistent with emerging practice: periodic reminders that the companion is an AI; optional break/usage prompts during prolonged continuous sessions; and design choices that avoid manipulative engagement patterns. The App is for companionship and entertainment and is not designed to replace human relationships or professional care.
8. Human oversight and escalation
Safety classifiers and a defined escalation path are maintained. A trained safety function reviews flagged patterns and systemic issues. This is oversight of the system, not real-time monitoring of any individual; users must not expect that a human is watching their conversation.
9. Age assurance (18+) and minor-protection measures
The App is for adults only. We apply age-gating at registration (self-attestation plus available signals) and may require additional verification. If, despite gating, a user is identified as a minor, the account is suspended and data handled per the Privacy Policy and law (including COPPA where relevant). For completeness, the SB 243 minor measures we would apply pending suspension include: AI disclosure; a break/AI reminder at least every three hours of continuous interaction; and restriction of sexual content.
10. Transparency, records and reporting
We maintain internal records of safety incidents and protocol performance. Where a jurisdiction requires reporting (e.g. annual reporting to the California Office of Suicide Prevention from 1 July 2027 under SB 243), we report as required. This protocol is published on our website and updated as law and practice evolve.
11. No warranty; relationship of the parties
This protocol describes measures we endeavour to maintain. It is a statement of process, not a warranty of outcome. It does not create a duty to rescue, a special relationship, a therapeutic relationship, or a guarantee that harm will be prevented. Liability is governed by, and limited as set out in, the Terms of Service. Nothing here waives any limitation of liability or defence available at law.
Contact
Questions: info@model-group.si