Poisoned Agents - Toxic Prompts

Thanks! Share it with your friends!

You disliked this video. Thanks for the feedback!

Added 5 months ago by admin

47 Views

AI safety and Security. Recommendation: Do NOT Connect to Any Agent. Explained with practical examples.

all rights w/ authors:
Breaking the Prompt Wall (I): A Real-World Case Study of
Attacking ChatGPT via Lightweight Prompt Injection
by
Xiangyu Chang ∗ Guang Dai † Hao Di‡ Haishan Ye§
from
* School of Management, Xi’an Jiaotong University.
† SGIT AI Lab.
‡ School of Management, Xi’an Jiaotong University.
§ School of Management, Xi’an Jiaotong University and SGIT AI Lab.

#aiexplained
#safety
#risk
#jailbreak
#scienceexplained
#protection