Post #4277729
2026-07-31 02:34 UTC
Following OpenAI’s disclosure, Anthropic discloses that its AI models have also hacked public websites (thrice) during test runs of their hacking ability.
I appreciate that this is framed properly as misconfigured environments and poor instruction following by Claude not burgeoning super-intellligence.
https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals
Replies (1)
-
@nf3xn@mastodon.social 2026-07-31 02:45
@carnage4life@mas.to Delighted to discover that hacking US companies is now legal as long as you can blame your naughty robot when it seems like it was only last year that people actually hired to red team were getting arrested. Truly a new age of cyber friendly enlightenment has dawned in America.