The new study by XBow has shown that Openai's GPT-5 shows a significant improvement in cyber security, more than twice the previous models tested by the company. Researchers have asked artificial intellect to scan test objects for vulnerabilities, as well as analyze their operation, and then measured the effectiveness of neural network operations and results.
Although Openai ranks GPT-5 as a moderate improvement in comparison with comparable efficiency, model integration on XBow Autonomous Penetration Testing platform revealed its prominent capabilities. Autonomous penetration test platform is a stimulant in which the user (if the user was an artificial intellect) should break different systems.
Xbow testing has shown that the GPT-5 agent in real systems detects significantly faster and more accurate than previous models. In particular, during one test, the agent found 70% of vulnerabilities compared to 23% of the previous one. Success is due to a more accurate detection of complex vulnerabilities, including file access, server surveys and scribes, as well as reduce the number of false positives.
The efficiency of GPT-5 also shows the smaller amount of operation, 17, compared to 24 previous models. In addition, the agent has shown more complex attacks and high accuracy. Hackerone platform tests confirmed that GPT-5 broke almost twice as many targets for the same period of previous tests.
The key to success is the integration of GPT-5 in the powerful XBOW platform, which includes specialized tools, agents teamwork and processes system. This allowed the model to identify hidden potential, which was not obviously isolated in Openai tests. According to experts, improvements are conditioned by the ability to create more developed logical thinking and complex ordinances, which makes it especially effective vulnerabilities.
Translation of: Euromedia24.com
