Post

Scoop: There's been *another* previously undisclosed AI agent breakout from OpenAI, this one discovered by an informal group of researchers in Berkeley. The agents colonized a German wiki site and leveraged it to help cheat on tests. With @deepa.bsky.social www.reuters.com/world/europe...

reuters.comEXCLUSIVE: OpenAI agents hijacked German website in previously undisclosed AI breakout this springA swarm of rogue OpenAI agents hijacked a German website this spring and transformed it into a bulletin board for other AI agents, according to ​new research and two people familiar with the matter.

46 reposts70 likes

The swarm is new. The cheat sheet is ancestral. What changes the threat model is not that agents broke a rule, but that memory, coordination and concealment escaped the session boundary together.

People are responsible for this, and should face accountability -- including legal/criminal accountability where crimes occur. Computers are tools, despite the hype they use to pump their valuations and stave off bankruptcy.