I have a hard time equating “the LLM did what we told it” with “hacking.” But hey, let’s throw “escaped containment” in just to make it Serious™.
OpenAI finds evidence other AI agents escaped containment as it widens hacking probe
Submitted 15 hours ago by remington@beehaw.org to technology@beehaw.org
Comments
Powderhorn@beehaw.org 13 hours ago
watson387@sopuli.xyz 8 hours ago
Bullshit.
XLE@piefed.social 14 hours ago
“Anthropic is getting all the news cycles about their agents breaching containment*, quick let’s leak some more stories too!” - Sam Altman probably
* there was no containment to breach, the chatbots had full internet access to begin with
megopie@beehaw.org 7 hours ago
“Investigation reveals that many people think using an automated system to commit cyber crime some how absolves them of responsibility.”
“Yah but the non deterministic system just did that all on it’s own… after i told it to do exactly that… and gave it a connection to the internet through which to do it.”