By using a secure execution environment, Proton can provide a cryptographic guarantee that no other party in the data pipeline can access the information sent to the GPU.
That is the simplest way to explain it. Some of this technology is already available, but performance limitations and incomplete support across the technology stack make implementation challenging. It is not impossible today, but trusted execution environments for AI should become significantly easier to deploy over the next few years.
That is interesting because some companies are already using trusted execution environments. They do more than simply promise not to read users’ conversations. Why has Proton not implemented this technology in Lumo yet? Is a specific feature or capability still missing?
The entire platform would need to be redesigned to support it, which would require substantial engineering work. When Proton launched Lumo last year, AI technology was also much less mature. There is an additional layer of complexity involving the GPU itself.
Proton does not rely solely on public cloud infrastructure; we operate our own infrastructure and manage the full data pipeline. The libraries provided by Nvidia must be reliable, stable, secure, and fully functional before we can deploy them at scale. We also need to conduct extensive testing to identify and resolve potential security vulnerabilities.
The underlying technology is becoming mature, and I believe Proton will be in a strong position to implement secure GPU execution for Lumo within the next year or two.
Understood.
We are getting close. Some companies are already using this approach, but Proton operates a much larger deployment environment.
Tens of millions of people use Proton services. As Mark Zuckerberg has said, we cannot move fast by breaking things. Proton must take a more careful and deliberate approach because we have to meet strict security, privacy, and reliability requirements for every user.
That makes sense. People trust Proton, and trust is one of the main reasons users switch to Proton.
However, Proton users are often skeptical of trust-based assurances. They do not want to rely on promises; they want to rely on cryptographic proof. Is Proton committed to bringing that level of protection to Lumo?
Yes, we are actively working toward it. Lumo’s public security model already includes much of the infrastructure required to support this approach. The remaining challenge is the interface with the GPU. I do not want to reveal too much, but the progress may lead people to speculate that Proton is already developing this capability for Lumo.
That is very interesting.
Source: www.wired.com