A blog from Tinfoil covering confidential computing and privacy-preserving AI infrastructure, with posts on how the company secures AI workloads and related technical developments.
Read / subscribe freeThis is the only entry from Tinfoil Blog in the hub so far.
Other resources that share this publication's topics.
A technical article covering API authentication and authorization mechanisms such as API keys, OAuth, and JWTs, comparing their trade-offs and describing common failure modes engineers encounter when implementing them.
An independent nonprofit offering research and programs on emerging technology topics including networked incitement, privacy and community safety, surveillance overreach, and digital harms that threaten democracy.
A Thinking On Paper podcast episode with Carissa Véliz, an Oxford associate professor working on AI ethics. She describes privacy as an ability, a right and a duty, explains what companies can infer from location data, and discusses algorithmic decisions that cannot be explained or appealed, decentralized identity, and choosing tools such as Signal and DuckDuckGo. Free to listen or watch.
An open-source, MIT-licensed SDK from Microsoft for running untrusted code — such as model output, plugins, and tools — in sandboxed containers on Windows, Linux, and macOS. It provides multiple containment backends, from OS-native process sandboxes to full virtual machines, behind a unified containment model, with policy-driven controls for filesystem, network, and UI access, and Rust, .NET, and Node SDKs.
Technical report examining whether off-grid solar microgrids in the United States could meet near-term AI datacenter power needs as a faster, lower-emission alternative to building new natural gas plants, based on 20-year powerflow modeling across thousands of site configurations.
A security writeup from Peter James: after asking Meta's Muse agent to archive the files it could see and send them to Google Drive, the download contained the root filesystem of the session's Linux environment — Muse's internal documentation, memory files, agent logs and SSH key files. The findings were reported through Meta's bug bounty program; the archive, keys and session logs were not published.
