Report: Off-the-Shelf OpenAI Model Uncovers Chrome Zero-Day

TL;DR: Forbes reported on October 13 that a general-purpose OpenAI model, with no special tuning, found a zero-day vulnerability in Google Chrome.

Key points

• According to the report, the model was a publicly available "off-the-shelf" model, not a purpose-built security tool.
• The target was Google Chrome, the world's most widely used browser. The flaw was reportedly a zero-day, meaning it had not been publicly disclosed.
• The original article has the details: which model was used, the type and severity of the bug, and whether it has been patched.

Why it matters

Finding zero-days has mostly been work for skilled security researchers using specialized tools. If general-purpose LLMs can do it too, defenders can find bugs faster, but attackers face a lower barrier as well. Browser patch cycles and bug bounty programs may need to adjust.

Read more

• An Off-The-Shelf OpenAI Model Found A Google Chrome Zero-Day Exploit — Forbes