LLM's and vulnerabilities

So the latest LLMs are better at exploiting one-day vulnerabilities according to a study. Much better at exploiting than finding. 87% success rate if the CVE has a description. That drops to 7% if there is no description.

GPT-4 nutzte 87 Prozent bereitgestellter One-Day-Vulnerabilities aus

Eine neue Studie zeigt auf, dass große Sprachmodelle (LLMs) wie GPT-4 One-Day-Vulnerabilities autonom ausnutzen können.

TARNKAPPE.INFOAntonia Frank

Excellent, just brill…

There’s going to be a better and quicker ROI for the nefarious folk before anyone else at this rate.

Just throw the link into a translator if you don’t have German. The study is in English

LLM Agents can Autonomously Exploit One-day Vulnerabilities

Mascot Sammy2023