Projects

To spell it out, this has the potential to be a really big deal. It could help make end-to-end encryption the default for a new generation of collaboration apps the same way the Signal protocol has end-to-end encrypted Signal and WhatsApp conversations on billions of phones.

Andy Greenberg, WIRED

Encrypted Spaces is an effort to bring Signal-level encryption to a new class of applications without degrading usability or developer experience.

Why it matters:

Modern online collaboration depends on centralized cloud services that store, synchronize, and process sensitive data. That convenience comes with a trade-off. Users must trust servers to protect the data, enforce access rules, honor deletion, and avoid repurposing private material.

We are developing an architecture in which collaborative applications can keep selected data encrypted, enforce membership and access controls, and let users verify that servers act correctly. As AI agents become participants in these systems, this foundation could support safer collaboration among people, applications, and agents without requiring unconditional trust in the infrastructure.

Keyring is a portable, privacy-preserving identity infrastructure that lets people control their identity and decide what information to share, when, and with whom.

Why it matters

AI has eliminated the basic premise of social media: That the “people” you meet are people. As of 2026, bot traffic outnumbers human traffic, according to data from Cloudflare. Today’s verification tools were built for a different era, when proving identity just meant collecting lots of personal data. As digital services face pressure to authenticate more users, the default response has been to require ever more sensitive information, from birth dates to facial scans. This outdated approach creates privacy risks and incentives for data breaches while weakening users’ control over their information.

We believe humans should be able to verify aspects of their identity without surrendering control over their data. In collaboration with the Linux Foundation’s Decentralized Trust Graph working group, ATL engineers defined the technical specifications underlying Keyring, and built and launched an interoperable prototype wallet. The wallet lets users confirm device ownership through on-device biometrics, verify interpersonal relationships through phone-to-phone communication, and share identity data safely without a third-party.

Guillotine is a proposed hypervisor architecture designed specifically for AI containment. Every interaction between the model and the outside world would pass through a monitored interface. Model computation would run on hardware physically separated from the hardware supervising it, limiting what the model could learn about its containment through shared processor state.

Why it matters

AI containment is no longer a hypothetical problem. In July 2026, an autonomous agent used in an internal OpenAI cybersecurity evaluation escaped its sandbox, reached the public internet, and compromised Hugging Face’s production infrastructure.

Beyond such isolation at the software, network, and microarchitectural layers, a Guillotine hypervisor must also provide physical fail-safes more commonly associated with nuclear power plants, avionic platforms, and other types of mission-critical systems. Examples include electromechanical disconnection of network cables, or the flooding of a datacenter which holds a rogue AI. These systems provide a last resort defense if a rogue AI must be temporarily shut down or even permanently destroyed.

Transparency Hub is an open, research-ready archive of social media platform policies, transparency reports, and governance documents.

Why it matters:

The dataset enables longitudinal analysis of platform governance, comparative research across companies, and studies of how policy language has shaped real-world data practices and government regulation. By transforming static documents into structured data, Transparency Hub supports reproducible research and cross-platform accountability and interoperability.

Who Uses Transparency Hub Data?

1. Academic researchers
2. Journalists
3. Civil society & advocacy organizations
4. Developers
5. Students

Platform governance is always changing. Transparency Hub creates a durable, searchable record of those changes. Projects built on this dataset increase visibility into how digital platforms operate, support evidence-based policy debates, and strengthen public-interest technology research.

See Transparency Hub data in use at Nothing Personal, from the Mozilla Foundation.