Agentic ProbLLMs: Exploiting AI Computer-Use and Coding Agents By Embrace The Red

Want to see some pro hackers decimate LLM’s “Security” layers as if they didn’t even exist?

I try not to just repost other people’s content, but this one was too good to pass up, sending you right to the source.

Allow me to introduce a brilliant presentation by Johann Rehberger from Embrace The Red, where he completely annihilates pretty much every mainstream coding agent.

It really is amazing watching a master at work, even if that work is obliterating any notion that security exists with these tools that everyone has been so quick to pick up.

I am not an absolutist when it comes to using AI tools, but I must say that this presentation makes most AI tools' security look like Swiss cheese - moldy Swiss cheese, with numerous holes in it.

So without further ado, please enjoy and let me know what you think about it.