Back to today’s briefing

Story perspectives

New Jailbreak Method Boosts Attack Success by 60%

1/4/2025

43 8

Summary

Story summary

A groundbreaking jailbreak technique named "Bad Likert Judge" has been uncovered by researchers at Palo Alto Networks' Unit 42. This method cleverly manipulates large language models (LLMs) to assess harmful content, boosting attack success rates by more than 60%. Urgent action is needed to enhance content filtering and protect against these emerging threats.

1 of 1

Story summary

A groundbreaking jailbreak technique named "Bad Likert Judge" has been uncovered by researchers at Palo Alto Networks' Unit 42. This method cleverly manipulates large language models (LLMs) to assess harmful content, boosting attack success rates by more than 60%. Urgent action is needed to enhance content filtering and protect against these emerging threats.