OpenAI presented an AI system that reproduced 1,724 Linux vulnerabilities during a 30-hour test, generating 948 automated tools for gaining root access and evidence for 291 information-disclosure cases.
Its five-stage workflow covers discovery, verification, assessment, patch generation and human review.
The results offer a concrete look at automated vulnerability research and the role of review in the process.
How the system works
The process followed by the system consists of five stages:
- Discovery: It identifies vulnerable points in the code.
- Verification: It confirms in practice that the security gap is real.
- Rating: It examines the consequences of emptiness.
- Create a correction: Compiles the necessary code to fix the problem.
- Human control: Researchers are examining the result before final disposal.
The importance of new technology
The system was able to identify at least 10 new attack techniques that were previously unknown. OpenAI emphasizes that automated patch generation is essential to protect systems from growing risks.
However, the use of artificial intelligence to identify security vulnerabilities has analysts concerned. The concern is that such tools can fall into the hands of people with bad intentions, allowing them to find and exploit vulnerabilities much more easily. It is indicative that the number of security vulnerability reports has more than doubled from January to August 2026.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
