1. Nature of the service Connection Safety AI is FRIC’s informational, privacy-conscious assistant. It synthesises approved FRIC material to help users notice warning signs, consider verification and pause before high-risk online requests. It assesses the interaction and supplied information—not a person’s character, identity or guilt. 2. Limits It cannot reliably verify identity, confirm fraud, predict conduct, investigate crime, recover money, provide a definitive risk score or guarantee safety. Responses may be incomplete, outdated or mistaken, particularly when information is limited, ambiguous or manipulated. 3. Human judgement Do not use an answer as the sole basis for reporting, blocking, disciplining, profiling, denying a service, making employment or credit decisions, or accusing someone. Institutional use requires proportionate human review, a documented purpose and a route to challenge or correct an assessment. 4. Sensitive information Do not enter passwords, one-time codes, banking or card details, cryptocurrency keys, full identity documents, live location, intimate images, health records or privileged communications. Remove identifiers where possible. Do not submit another person’s private messages without an appropriate lawful basis. 5. Urgent support For immediate danger, coercion, self-harm risk or threats, contact local emergency or safeguarding services. For transferred funds, contact the bank or payment provider promptly. For account compromise, use the official service to change credentials and activate multi-factor authentication. 6. Privacy and model use FRIC applies minimisation and user-control principles. Private messages are not used to train general-purpose models by default. A production deployment must explain model providers, processing, retention and deletion through its notice, contract and consent interface. 7. Fairness and feedback Risk guidance should focus on observable interaction patterns—not race, nationality, religion, gender, disability or other protected traits. Users should be able to report, question or seek correction of an answer and request human support. 8. User acknowledgement By using the assistant, you acknowledge these limitations and agree not to use its output as proof of wrongdoing.