Machines Like Us In the Wake of Tumbler Ridge, Can We Trade Privacy for Safety?
Mar 10, 2026
Meredith Whittaker, President of Signal and long-time privacy advocate, discusses end-to-end encryption and privacy-preserving AI. She explores risks of mandatory message scanning, how surveillance-driven tools erode rights, and why agent-style AI threatens app-level encryption. Conversations cover age verification, surveillance incentives, and reframing privacy as protection for relationships and dissent.
AI Snips
Chapters
Transcript
Episode notes
AI's Architecture Makes It Lean Toward Surveillance
- Current deep learning AI is materially tied to surveillance because models scale with massive datasets and compute, favoring incumbents with cloud infrastructure.
- Whittaker explains why winners of early web platforms became dominant AI players through data and compute economies.
Politically Define AI Goals Before Building The Tools
- Decide collectively what AI should do and fund alternatives that align with public goals instead of leaving choices to big tech.
- Whittaker urges political debate on desired social outcomes (schools, pedagogy) before tech design and governance.
AI Agents Break App-Level Privacy By Design
- Agents require broad access to user data plus the ability to act autonomously, which creates new attack surfaces that can bypass app-level encryption.
- Whittaker gives the example of an agent needing calendar, browser, card, and Signal access to plan events on your behalf.