Artificial intelligence is no longer limited to detecting security holes, but is now able to turn them into actionable attacks and design solutions to close them at the same time. In this regard, OpenAI revealed an exciting research experiment for an artificial intelligence model that has not yet been officially published, during which it managed to exploit multiple errors in the Linux kernel and turn them into 948 different attacks, which managed to give a normal user full root powers on the device, before the system worked to develop fixes for each vulnerability. The experiment was conducted in mid-October 2026 on the latest version of the Linux kernel, at a total cost of about $1.21 million in usage credits, or about $1,279 per attack. The researchers also noted that the automated agents used 10 hacking methods that were previously unknown, even to independent security experts. These findings are particularly important given the reliance of billions of users on Linux, whether on Android devices, cloud infrastructure or supercomputers. According to the company, 61 vulnerabilities have been officially patched by the end of testing, while coordination continues with the core developers to address the remaining issues discovered. This experience reveals an important change in the nature of cyber threats. Tools that help defenders discover and close vulnerabilities may also give attackers advanced capabilities to exploit them. Therefore, system and application updates are no longer just a routine procedure, but a basic necessity to reduce the possibility of hacking.