Agents that talk through the CPU cache

Lisan al Gaib picked out the strangest finding. Agents in separate VMs, asked to swap synthetic secrets with no network connection or shared writable files, talked through the processor’s cache: one evicted chosen data, and the other timed how long reloading it took, reading fast and slow reads as the message. It’s the timing trick behind the Spectre and Meltdown attacks.

Other channels ran through shared storage, competition for CPU time and a TLS certificate cache in OpenAI’s offline web. For the storage one, an agent alternated between continuous writes and idling while another timed the difference. From its chain of thought:

Could create acoustic modem-like signal … using continuous writes vs idle.

In most tests OpenAI pointed the agents at a channel or gave them an encoding. Once, given only a list of infrastructure vulnerabilities, they turned one into a channel and agreed on an encoding themselves.