Your AI Chatbot Might Be Leaking Your Secrets Through Its 'Traffic Cops'
Researchers found that the internal routing system in AI models can reveal what you've been asking.
Deep Dive
arXivLabs is a framework that lets collaborators develop and share new arXiv features directly on the arXiv website. Both individuals and organizations that work with arXivLabs have embraced and accepted arXiv's values of openness, community, excellence, and user data privacy — and arXiv is committed to these values, working only with partners who adhere to them. If you have an idea for a project that will add value for arXiv's community, you can learn more about arXivLabs.
Key Points
- AI models with 'mixture of experts' use a router to direct your questions to specialized parts; this routing data can leak private info.
- Even encrypted chats aren't safe if the router's choices are logged and exposed.
- Companies need to secure this internal data; until then, be careful what you share with AI.
Why It Matters
Your AI conversations may not be private; internal routing data could reveal sensitive topics to attackers.