Making Tech Headlines · Page 9

OpenAI's Cyber Model Finds Hundreds of Ways Into Linux, Then Hands Humans the List

November 7, 2026 at 6:01 AM EST · New York time

OpenAI says its GPT-6 Cyber model helped an automated system identify 1,724 Linux kernel security flaws and exploit 948 to gain administrator-level access during testing. At least 10 techniques were previously unknown, according to the company. The testing used about $1.2 million in API credits, and human researchers reviewed the findings before vulnerabilities were reported. The results show both the scale of the company's claimed testing and the continuing role of people in checking what the system found. The work offers an appealing pitch for automated security: software can examine complicated code and find ways around its defenses. It also shows why powerful security testing attracts careful attention. A system that demonstrates how to gain access is producing information that must be handled responsibly. The tireless auditor has produced a very long list. Its human colleagues still have the less glamorous job of deciding what to do with it.

Satire. Not factual reporting. Read the original reporting from the linked publishers.