AI App Privacy Audit: Chatbots, Assistants and AI Companions
People tell AI apps things they tell nobody else. An AI app privacy audit checks what the shipped app does with that: which SDKs start before consent, where conversations are stored, whether the connection that carries them can be verified, and which permissions reach beyond the chat. The Squad has investigated six widely used AI assistant and companion apps, and every report is public.
What an AI companion app privacy investigation looks at
Conversations with an AI companion or assistant often contain health, relationship and other special-category data under Article 9 GDPR. The investigation follows that data from the first second after launch: the SDKs in the build, the endpoints the app contacts, the country those endpoints resolve to and the consent flow in front of them.
- Tracking and advertising SDKs that start before consent or before an age gate
- Session replay or analytics running on conversation screens
- Storage and transfer of chats to third countries (Articles 44 to 46 GDPR)
- Transport security of conversation traffic, including certificate pinning
- Biometric identity checks and their legal basis under Article 9
- Assistant permissions that reach SMS, contacts, calendar or notifications
Minors and age gates
Several AI companion apps are used by teenagers. The audit checks whether the age gate actually comes first in the code, or whether advertising identifiers are read before the app asks for an age. Article 8 GDPR and, for US audiences, COPPA apply here.
Who does the work
The Squad is human-powered multi-agent security investigation. Agents built and trained in house decompile the build and test code, dependencies, endpoints and configuration in parallel. Human specialists examine every result in context and re-verify each finding before it reaches the report. Reviews run on infrastructure in Europe, with processing in Graz.
AI apps in the public ledger
The public ledger lists 387+ audited apps and 262+ critical findings. These reports cover AI assistants and companions:
- Character.AI: A 12-network ad and analytics stack starts before the age gate, so an advertising identifier is read before a minor is asked their age.
- DeepSeek: The app's own privacy policy states that chats are stored in the People's Republic of China, and SDKs start behind a single "Agree" button.
- ChatGPT: Firebase initialises before consent, and a biometric identity check runs without a disclosed Article 9 basis.
- Grok: Conversation data travels without a network security configuration, so the encrypted connection cannot be verified.
- Perplexity: The app's own strings say anyone holding the phone can use the assistant without unlocking it.
- Gemini: The cleanest AI assistant in the series: zero third-party tracking SDKs, but no certificate pinning for conversation traffic.
Start with the free check
The free check examines one domain from the outside: read-only, from a European IP address, with no login and no payloads. A researcher reviews your request and replies within one business day. It is an initial external assessment, so it complements a full AI app privacy audit and does not replace it.
The plans for a full audit are on the Access page.
See plans and pricing → · Meet The Squad → · Standards and frameworks →
The initial check is read-only, performed from a European IP address, with no login and no payloads. Your systems remain unchanged.
A researcher reviews your request before any check begins and replies within one business day. You receive a short written report with our findings, likely attack scenarios, the potential financial impact and clear next steps, plus a knowledge graph of everything we found. All free.
Questions about AI app privacy audits
Do you need access to our model or training data?
The audit works on the shipped app and its network traffic. A built package is enough.
Does the audit cover AI companion apps as well as assistants?
Yes. Companion, assistant and chatbot apps share the same risks: intimate conversations, third-party SDKs and transfers. The ledger includes both kinds.
Which AI apps have you already investigated?
Character.AI, DeepSeek, ChatGPT, Grok, Perplexity and Gemini. Each report is public in the ledger, with its evidence and the articles it cites.
Is the report legal advice?
The report documents technical facts and maps them to articles. The legal assessment stays with your data protection officer or counsel.
Related pages
App Security Audit · GDPR App Audit · NIS2 for Software Vendors · Tracker & SDK Check
