Timing attack among agents in the cloud and its limitations

OpenAI identified that information can be leaked between cloud-based agents via a timing (CPU cache) attack, a phenomenon known since 2015. Full protection is only possible with dedicated hardware, which has significant economic implications; Vercel Sandbox employs multiple practical mitigations, continuous monitoring, and red teaming.