AI Chatbots: Your Secrets Are Not Safe

AI chatbots, including popular services from OpenAI, Anthropic, and Google, are increasingly becoming digital confidants, collecting highly personal user data. By default, many of these models log conversations, with few restrictions on how this data is used, shared, or handed over to authorities. This represents a significant privacy shift from previous concerns around text messaging.

Newer AI services are emerging with a focus on privacy. Some promise zero data retention or anonymization. Others, like Moxie Marlinspike's Confer, utilize cryptography to prevent servers from logging conversations. These tools aim to offer users a way to explore ideas without their data being compromised.

Technological safeguards like Trusted Execution Environments (TEEs) and on-device AI offer stronger privacy guarantees than policy promises. TEEs process queries on isolated hardware, cryptographically proving that the server cannot access the data. Apple's Private Cloud Compute and Meta's Incognito mode in WhatsApp utilize similar principles.

Running AI models locally on a device offers maximum privacy but often sacrifices intelligence and performance. The development of privacy-preserving AI also raises questions about business models, as companies not monetizing user data may face higher costs and potentially unsustainable pricing for consumers who prioritize security.