Jailbreak Anthropic’s new AI safety system for a $15,000 reward

Posted by:

|

On:

|

In testing, the technique helped Claude block 95% of jailbreak attempts. But the process still needs more ‘real-world’ red-teaming.

Posted by

in