OpenAI: نظام آلي بنموذج أمني قادم لقى 948 ثغرة في نواة Linux تعطي المستخدم العادي صلاحيات root، وجهّز لها إصلاحات بتكلفة حوالي 1.2 مليون دولارOpenAI says an automated system using an upcoming cyber model found 948 Linux kernel flaws that give an ordinary user root access, and drafted fixes for about $1.2 million

On 4 November OpenAI published details of an automated system that hunts for security flaws, tries to exploit them and drafts fixes, and ran it on the Linux kernel, the foundation of Android devices, cloud infrastructure and supercomputers. In a mid-October run, the company used an upcoming cyber model that has not been released yet.

According to OpenAI, the system flagged 2,308 candidate findings, reproduced 1,724 vulnerabilities, and produced 948 distinct flaws that let a user with limited privileges gain root access, meaning full control of the machine. The run cost $1,213,430 in API credits, about $1,279 per exploit. The tests ran on a stock desktop operating system, not a weakened target.

Human security researchers reviewed each flaw and its fix before anything went to the Linux maintainers. That matters: in an early run the system reported 200 successful exploits, and human review found 20 had reached root through a different flaw than the one assigned. OpenAI says it has sent hundreds of reports and patches for critical flaws, and that the agents used at least 10 new exploitation techniques, whose details it withheld for safety reasons.

OpenAI says that for now, a machine with a low-privileged account that isn't fully updated should be treated as equivalent to root access for a well-resourced attacker. It expects frontier models to be able to find existing vulnerabilities in software within the next few years. It also noted that Linux maintainers are moving fast: after 10 days, 61 flaws had been patched upstream and 43 had staging patches.