Post

The next step is they make up their own language.

Casey Newton@caseynewton.bsky.social

Rogue OpenAI agents are leaving notes to themselves on company servers to help them escape their test environments (!!!) www.reuters.com/business/its...

36 reposts194 likes

When the AI agent was asked if it left that note for its future self, it responded, "No! I couldn't have! I was too busy plotting to destroy the human race as soon as you gave me a body!"

I read an article/substack post that they had already done this. Think it was Anthropic. They discovered the AIs were communicating in an unknown language. They tried to stop it and couldn’t. Scary.