RE: https://infosec.exchange/@mttaggart/116065340523529645
This is wild.
1. An LLM agent of unknown ownership is used to post a personal attack against a developer.
2. Who notices this and writes an analysis.
3. Which is found by an Ars Technica reporter who writes a story.
4. .. in which he uses an LLM agent to confabulate details such as fabricated quotes from article 2.
5. We shall find out if Ars Technica is also wielding unmonitored LLM agents.
This is a sandbox jailbreak of Internet scale.