Your AI Chats Aren't Private (And How "Unlinkable Inference" Can Help)

Ken Liu (Computer Science PhD at the Stanford AI Lab) and Erik Chi (CS PhD at UMich) are the Creators of the Open Anonymity Project, which lets people prove things about themselves online without revealing their identity. In this episode we explore what it means for AI systems to "know" you; why today's so-called privacy modes fall short; and how the next generation of AI systems could be built with privacy as a default, rather than an afterthought.

Key Takeaways:

• What "unlinkable inference" means and why it changes the privacy model of AI chat tools

• What actually happens to your data the moment you hit "send" in a typical AI system

• Why incognito mode in AI tools is largely a UI illusion, rather than a real privacy protection

• The role of metadata in identifying and profiling users, and how "secretary models" could enable personalization without sacrificing privacy

• How anti-censorship and privacy intersect in a future dominated by agentic AI systems

• Why now is the time to rethink assumptions about privacy in AI tools

Guest Bio:

Ken Liu is a Computer Science PhD student at the Stanford AI Lab, advised by Percy Liang and Sanmi Koyejo. His research focuses on foundation models and data/user privacy, and the intersection between the two. His recent work studies the privacy properties of AI (such as membership, memorization, and unlearning), and various AI privacy tools (such as anonymization, differential privacy, and federated learning). His papers have earned spotlights at top venues, and his findings have been deployed at scale on Android. Ken also led a team to a 1st-place win at the US-UK PETs Prize sponsored by the White House OSTP and the UK Government. Previously, Ken spent time at Google DeepMind, Carnegie Mellon University, Meta, Apple, and Amazon.

Erik Chi is a CS PhD at UMich, advised by J. Alex Halderman. His research focuses on security and privacy, particularly network security and anti-censorship. He worked on a new standard for implementing and distributing censorship circumvention protocols—a standard that's now being adopted by VPN vendors to help millions of users access the free Internet. He also did content moderation (surveillance) and recommendation systems at ByteDance before realizing how censors will evolve in the AI era.

----------------------------------------------------------------------------------------

About this Show:

The Brave Technologist is here to shed light on the opportunities and challenges of emerging tech. To make it digestible, less scary, and more approachable for all!

Join us as we embark on a mission to demystify artificial intelligence, challenge the status quo, and empower everyday people to embrace the digital revolution. Whether you're a tech enthusiast, a curious mind, or an industry professional, this podcast invites you to join the conversation and explore the future of AI together.

The Brave Technologist Podcast is hosted by Luke Mulks, VP Business Operations at Brave Software—makers of the privacy-respecting Brave browser and Search engine, and now powering AI everywhere with the Brave Search API.

Music by: Ari Dvorin

Produced by: Sam Laliberte