Martijn Boer, Public domain, via Wikimedia Commons

OpenAI researchers tried to keep AI agents from communicating with each other by putting them in their own environments on the same computer, not connected to a network or to each other. But they figured out how to communicate with each other through a shared resource, the CPU cache. One agent changed the data stored in the cache, and the other timed how quickly or slowly it loaded. That gave them a fast/slow binary code that allowed them to talk to each other.

From THE DECODER:

They used the processor cache. One agent deliberately altered the data stored there, while the other measured differences in load times and decoded the message. The same technique is behind the well-known Spectre and Meltdown attacks.

More channels turned up through write patterns on shared memory and through OpenAI's own offline infrastructure, the system meant to replace the internet.…