If you discover a security vulnerability in Fud AI, please report it privately so it can be addressed before public disclosure.

Preferred: Use GitHub's private vulnerability reporting — it's end-to-end private and routes directly to the maintainer.

Alternative: Email either of:

• apoorv@fud-ai.app
• ad13dtu@gmail.com

Please include:

• A description of the vulnerability
• Steps to reproduce (a minimal PoC is ideal)
• Affected platform + version (iOS — check Settings → About, or MARKETING_VERSION in ios/calorietracker.xcodeproj/project.pbxproj; Android — check About screen, or versionName in android/app/build.gradle.kts)
• Any potential impact — data exposure, API-key leakage, code execution, etc.
• Your name / handle if you want credit in the release notes (optional)

You can expect an initial acknowledgement within 7 days, and a more detailed response (triage + estimated fix timeline) within 14 days. Please do not disclose the issue publicly until a fix has shipped to the App Store and Play Store.

Only the latest released version on each store (App Store for iOS, Play Store for Android) is supported with security updates. The repository's main branch tracks the next release for both clients.

In scope (iOS):

• The iOS app source in ios/ (SwiftUI codebase, widget extension, tests targets)
• API-key handling and iOS Keychain storage (KeychainHelper, AIProviderSettings, SpeechSettings)
• Network requests to AI and speech-to-text providers (GeminiService, ChatService, SpeechService), including multi-image food analysis and expanded nutrient payloads
• Fud AI Plus proxy request handling (web/api/gemini.js), including Gemini food/Coach routing, Deepgram voice routing, anonymous install IDs, and quota enforcement
• Barcode lookup behavior against Open Food Facts, including the "missing product / missing nutrition" fallback path
• HealthKit read/write paths (HealthKitManager) and UUID-tagged sample conventions
• Widget App Group container (group.com.apoorvdarshan.calorietracker) and the snapshot written into it
• Local persistence layer (UserDefaults, Keychain) including the Coach chat history and food/weight/body-fat logs

In scope (Android):

• The Android app source in android/ (Kotlin + Compose codebase, Glance widget, repositories, services)
• API-key handling via EncryptedSharedPreferences (AES-256, AndroidKeystore-backed) in data/KeyStore.kt — including the AEAD recovery path that wipes a corrupted master-key alias on reinstall
• Network requests to AI and speech-to-text providers (services/ai/*, services/speech/*), including multi-image food analysis, Coach image attachments, and expanded nutrient payloads
• Barcode lookup behavior against Open Food Facts, including the "missing product / missing nutrition" fallback path
• Health Connect read/write (services/health/HealthConnectManager.kt) and the fudai_<uuid> clientRecordId convention used for dedup + safe deletion
• Glance widget snapshot (models/WidgetSnapshot.kt) shared via the app's DataStore and ImageProvider(bitmap) rendering path, plus local thumbnail caching used by food rows and saved meals
• Local persistence (DataStore Preferences for everything, EncryptedSharedPreferences for keys, no Room/cloud)
• Release signing material handling — keystore.properties and *.jks are gitignored; ProGuard/R8 keep rules in app/proguard-rules.pro (relevant if a missing keep introduces a release-only crash that has security implications)

In scope (web):

• The marketing site source in web/ (static HTML/CSS, no JS framework, no backend)
• The Fud AI Plus serverless API route in web/api/gemini.js, including required secret handling for GEMINI_API_KEY, DEEPGRAM_API_KEY, and optional KV quota storage
• Privacy policy + terms pages (web/privacy.html, web/terms.html) — accuracy of disclosures

Out of scope:

• Vulnerabilities in third-party AI providers (report to them directly — OpenAI, Anthropic, Google, xAI, etc.)
• Vulnerabilities in third-party speech-to-text providers (report to them — Deepgram, AssemblyAI, etc.)
• Incorrect or incomplete Open Food Facts product data. Fud AI can only show the nutrition available from that public database; missing products are product-data issues, not security issues
• Vulnerabilities in OS-level components (iOS HealthKit, Android Health Connect, Glance, WorkManager, Compose) — report to Apple / Google
• Issues requiring physical device access with the device unlocked
• Social-engineering attacks against users' own API keys
• Denial-of-service against the user's own AI provider via API quota exhaustion (that's a user-controlled cost, not a security boundary)

If you make a good-faith effort to comply with this policy during security research, we will not pursue or support any legal action related to your research. Please don't access or modify user data, avoid service disruption, and give us reasonable time to fix issues before disclosure.

Researchers who responsibly disclose valid vulnerabilities will be credited in the release notes and in the commit message that ships the fix (unless they request to remain anonymous).