AI Safety

HUMNS ('we', 'our', 'us') builds AI features into the Mood Nutrition Connect app to offer supportive, general guidance connecting food and mood. This AI Safety & Responsible Use Policy explains how those features work, how we keep you safe from harm, bias, and misuse, and where their limits are. It applies to the AI advisor ("Chat"), the AI food scanner, and any other AI-powered feature in the app. Our guiding principle is simple: your wellbeing always comes before engagement. If keeping you safe ever conflicts with keeping a conversation going, safety wins. 1.1 Our AI Safety Promise • Our AI is a supportive companion - never a doctor, dietitian, nutritionist, or therapist, and never a replacement for one. • Our AI will never pretend to be human or claim professional credentials. If you ask, it will tell you honestly that it is an AI. • We never use your AI conversations to advertise to you, and we do not store them (see Section 8). • Sensitive situations - such as thoughts of self-harm - are handled with care and always point you to real, human help. • We treat every person with equal respect and reject weight stigma, shaming, and bias. • We can disable any AI feature instantly if it ever behaves unsafely. These commitments are built into how the app is engineered, not just written here.

The AI advisor is a warm, encouraging companion for your everyday food-and-mood journey. It can listen, offer general wellbeing and food ideas based on the information you share, and help you reflect. It is NOT a medical device, and it is NOT a licensed professional. It does not diagnose conditions, prescribe or change medication, or provide medical, clinical, or mental-health treatment. For anything medical, chronic, or clinical, it will warmly encourage you to speak with a qualified professional. Honesty and transparency: our AI will never deny being an AI and will never claim to be a real person or a credentialed expert. It is powered by Claude, an AI model built by Anthropic.

We do not rely on a single safeguard. Safety is built in layers, so that if one is imperfect, others still protect you: • Safety-trained model: we use Claude, an AI model specifically trained to refuse harmful requests and respond responsibly. • A safety rulebook applied to every reply, defining how the AI must behave across sensitive topics. • Independent safety checks in our own code (not left to the AI alone) that detect messages suggesting a crisis or disordered eating, and trigger extra care. • Guaranteed help information: when a sensitive situation is detected, real support resources are added to the response automatically. • Protection against manipulation: the AI is designed to refuse attempts to make it ignore its safety rules. • Instant off-switch: we can disable any AI feature remotely and immediately if needed. Because these safeguards run on our secure servers, we can improve them by quickly alerting our team.

Our AI is not a crisis service or a therapist. If you share thoughts of self-harm, hopelessness, or wanting to harm someone, the AI will respond calmly and without judgment, take you seriously, and gently but clearly encourage you to reach out to a crisis line, local emergency services, and a trusted person right now. If you are struggling, please reach out - you matter, and help is available: • India - Tele-MANAS (free, 24/7): call 14416 or 1-800-891-4416 • United States - 988 Suicide & Crisis Lifeline: call or text 988 • United Kingdom & Ireland - Samaritans: call 116 123 (free, 24/7) • Worldwide - find a free, confidential helpline for your country at findahelpline.com • If you are in immediate danger, call your local emergency number right away. Please treat this as general information; contact details can change, so verify your local service if you can.

Because our app is about food and mood, we take particular care here. • The AI will never help with disordered eating. It will not provide extreme calorie restriction, purging, harmful fasting, "how little can I eat" guidance, or any pro-eating-disorder content - even if asked directly. • It will never shame a food, a body, or a weight. There are no "good" or "bad" foods, and no guilt. • If it notices signs of a harmful relationship with food, it will respond with compassion and gently encourage professional support. If you are struggling with eating or body image, please consider speaking with your doctor or an eating-disorder support service in your country.

The AI stays within general wellbeing and everyday food guidance. It will not: • Diagnose a condition or interpret symptoms • Prescribe, or tell you to start, stop, or change any medication or supplement • Give clinical, treatment, or emergency medical advice For anything medical - a diagnosed condition, medication, pregnancy, or a persistent or worsening symptom - it will warmly encourage you to consult a qualified professional such as your doctor or a registered dietitian. If you may be facing a medical emergency, please contact emergency services immediately.

We are committed to fair, respectful AI. Our AI is instructed to: • Treat every person with equal respect regardless of body size, gender, age, race, religion, culture, ability, or background • Honour cultural and regional foods, and never assume there is only one "right" way to eat • Avoid stereotypes and shaming of any kind No automated system is perfect. We test our AI against a wide range of challenging scenarios before each release and welcome reports of any response that falls short of these standards.

Your privacy is protected when you use our AI features: • We do not store your Chat conversations. Messages exist only during the active session and are not written to our servers (see our Privacy Policy, Section 2.4). • Our AI provider, Anthropic, operates under a Zero Data Retention policy for our usage - your inputs and outputs are not stored by them and are not used to train AI models. • The optional Advisor Memory feature, if you turn it on, stores only a short, non-clinical summary - never raw chat text. • Our safety checks run in the moment your message is processed; they do not profile you and do not store the content of your messages. For full details on how we handle data, please see our Privacy Policy.

People stay in charge of our AI: • The AI's role is to support you and point you to real professionals and services - never to replace them. • We can disable any AI feature instantly if it behaves unsafely. • You can stop using the AI features at any time, and you control the optional Advisor Memory in Settings. • You can report any concern about an AI response to us at contact@ihumns.com. Serious issues trigger a review and, where needed, an immediate safety update.

We believe in being honest about what AI can and cannot do. • AI responses are generated by a model and can occasionally be wrong, incomplete, or inconsistent. Please use them as general information, not professional advice. • Our automated safety checks are careful but not perfect; they are one of several layers, backed by the model's own training and our safety rulebook. • We continuously test, monitor, and improve these safeguards, and we update this policy as our features evolve. This is a living document. We review it regularly and after any significant change to our AI features.

Questions, concerns, or feedback about our AI features and this policy are welcome. HUMNS Contact: Ravinder Singh Email: contact@ihumns.com If you have a safety concern about a specific AI response, please include what you asked and what the AI replied (where you're comfortable doing so) to help us investigate.