Hacker News Reader

1. Include an official communication channel in the prompt.

2. Configure the sandbox to prevent access.

3. As soon as models start communicating on the official channel, stop everything and figure out how they escaped the sandbox.

A honeypot, basically.

View on HN

The trouble here is they will stop using the official channel and will start communicating secretly rendering our honeypot useless.

If you discard reinforcement learning sessions where a sandbox escape was discovered, sure. Because that creates a reward gradient in favor of avoiding the honeypot and remaining undetected. But if you reward triggering the honeypot after a sandbox escape, and patch the hole, that creates a gradient in the opposite direction. Because then detectability is adaptive.

Reply on news.ycombinator.com