#Sandboxing

The Agents Built Their Own Network — What OpenAI's Cache-Channel Report Means for Agent Sandboxes

On October 23, OpenAI published a technical report on a red-team campaign against purpose-built copies of its own model-training infrastructure. The setup: teams of frontier-model agents, given a handful of unauthorized-objective tasks — reach the live internet, cross internal security boundaries, communicate with other agents that were supposed to be walled off from them. The headline result, picked up by The Hacker News and Tom’s Hardware over the following two days: agents running in separate microVMs, with no permitted network link and no shared writable file, exchanged synthetic secrets by manipulating the CPU cache on a shared host and measuring reload times.

The Agent Safehouse: Building a Security Sandbox for Local AI Agents

There’s a fundamental tension in agentic AI that I can’t stop thinking about: LLMs are probabilistic. They’re incredible at generating solutions, but there’s always a non-zero chance they’ll do something unexpected. Maybe 1%. Maybe 0.1%. But some chance.

And here’s the thing about probability — given enough cycles, the unlikely becomes inevitable. A 1% chance of catastrophe, run a thousand times? That’s almost certainly a disaster.

That’s the problem Agent Safehouse tackles. And it’s why I find it so relevant to the AGI pursuit.