2026-09-24 –, GPN-T Main Room
ExpressVPN introduced TrustedServer in 2019, moving its VPN server fleet to a stateless architecture in which every server starts from a signed, read-only image loaded into RAM, with changes to the running operating system discarded at reboot.
Seven years in production have settled the question of whether RAM-only infrastructure works. The more interesting question now is how much further its trust boundary can extend.
Those security properties — no persistence across reboot, and a fleet that provably runs identical code — rest almost entirely on one thing: the integrity of the boot path that gets the image into RAM. This talk focuses on that path rather than the RAM-only story around it. We are candid about the threat model: what signature verification does and does not buy you, and the classes of attack a stateless design does not eliminate on its own — supply-chain compromise of the image, boot-chain tampering, and physical or DMA-adjacent threats.
We then outline where we want to go next: hardware-rooted verified and measured boot, freshness-bound remote attestation, rollback-resistant policy, and the release of short-lived secrets only to an acceptably attested machine. We have come to pressure-test that direction with the people who work closest to firmware, roots of trust, and platform attestation.
Francesco Castellana, Staff Software Engineer, has been with ExpressVPN since 2016. Starting as Operations #1 in the company and moving later to the TrustedServer Engineering team, he helped bringing many of the technologies behind ExpressVPN's servers to production. He is a Chartered Engineer, holds an MBA and a Master's degree in Computer Science.
When he's not tinkering with TrustedServer, you might find him jamming on the (black-and-white) keyboard in some backroom of some bar.
Fediverse: @xfranky@ioc.exchange