Post #4285039
2026-07-31 15:58 UTC
Just to be clear, the Anthropic LLM was instructed to behave as if it were sandbox, and it did. But Anthropic was just kidding. It had direct internet access. Whoopsie daisy.
The amount of YOLO in all of this is completely terrifying.
Replies (1)
-
@wdormann@infosec.exchange 2026-08-01 03:43
Not to be outdone by Anthropic's felony bragging, OpenAI assures us that they also did more felonies. https://www.reuters.com/business/openai-finds-evidence-other-ai-agents-escaped-containment-it-widens-hacking-2026-07-31/