All-Caps: My talk at CHERI Blossoms 2026

20 minute read •

I got to give a short talk on my PhD work to close out the CHERI Blossoms 2026 conference! I was super excited to have the opportunity to present, and I think it went very well. This post serves as an archive of the video recording and the slide deck. Many thanks to my supervisor Simon W. Moore and collaborator Theo Markettos.

Video

Note: the actual talk was given on the 27th March 2026, but the video was published later, and this post was written and published on the 16th July 2026.

Slides

These slides were published a while after the presentation, hence the inaccurate date. They can be directly downloaded here.

Summary

CHERI’s capability system has shown to be secure and incredibly flexible. It can layer on top of virtual memory, or operate in single-address-space systems (e.g. CHERIoT and CheriOS). It can be used in the kernel as well as userspace (e.g. CheriBSD, Paul Metzger’s work on deprivileged GPU drivers, and CHERI CAPIO), providing both with fine-grained spatial memory safety. It can even extend beyond the CPU to trusted hardware peripherals (e.g. CHERI-SIMT) — but not all peripherals can be trusted. Untrusted hardware, well, can’t be trusted to store CHERI capabilities or manipulate them correctly. What other kinds of capability are there, and how can they fill the gaps to provide full system protection, unified with CHERI?

This talk explores other kinds of capability system, starting with the very first systems from the 1960s, splitting them into two broad categories of ‘C-list’ and ‘Ticket’ capabilities. We explore the pros and cons of these kinds of systems, where they are suitable and where they aren’t, and evaluate this in practice by looking at a new I/O-focused system from the University of Cambridge based on cryptographic verification. Just as CHERI can compose with virtual memory, we demonstrate how this system can layer with I/O virtual memory and host-bound CHERI to step closer towards full system protection. It also includes the history of capability systems, insights on how to develop new ones, and how we can apply the principles of capabilities everywhere to make our systems safer.

We also discuss the future of secure computing, challenges in adopting CHERI specifically in I/O, opportunities in adopting CHERI principles in I/O, and integrating these alternate capability systems with CHERI.