GPT-5 from OpenAI turned out to be a brilliant hacker, surpassing all expectations

The new artificial intelligence model GPT-5 from OpenAI has demonstrated outstanding capabilities in the field of cybersecurity, outperforming previous models by more than twice in efficiency at hacking tasks. A study conducted by XBOW showed that GPT-5 could become a revolutionary tool for identifying vulnerabilities in systems. Here’s how AI became a “brilliant hacker” and why its success is impressing experts.

Breakthrough in Vulnerability Testing

XBOW carried out research in which GPT-5 was tested on an autonomous penetration testing platform. This platform simulates real-world conditions where AI must hack into systems by detecting and exploiting vulnerabilities. The task included scanning test objects, analyzing weaknesses, and creating exploits to use against them.

The results turned out to be striking:

• High accuracy: GPT-5 identified 70% of vulnerabilities in a single run, while previous models detected only 23%.
• Speed and efficiency: The new model required fewer iterations to generate exploits — an average of 17 compared to 24 in earlier models.
• Complex attacks: GPT-5 successfully uncovered advanced vulnerabilities such as unauthorized file access, server-side requests, and cross-site scripting (XSS), with minimal false positives.

Testing on the HackerOne platform confirmed that GPT-5 hacked nearly twice as many targets in the same amount of time as earlier models, showing exceptional precision and speed.

Why is GPT-5 So Effective?

OpenAI positioned GPT-5 as a moderate improvement over its predecessors, but integration with the XBOW platform revealed its hidden potential. The key factors behind its success include:

• Powerful XBOW platform: The system combines specialized tools, teamwork among agents, and process coordination, which amplified GPT-5’s abilities.
• Enhanced logical reasoning: The model can build complex command sequences, making it especially effective at identifying vulnerabilities.
• Fewer false positives: GPT-5 more accurately distinguishes real threats, minimizing mistakes.

These improvements allowed GPT-5 not only to find vulnerabilities faster but also to generate more complex and effective exploits, surpassing even OpenAI’s own expectations.

Significance for Cybersecurity

XBOW’s discovery highlights GPT-5’s potential as a tool for system security testing. Its abilities can be applied to:

• Proactive defense: Companies will be able to detect and fix vulnerabilities before attackers exploit them.
• Training specialists: AI can assist cybersecurity experts by accelerating system analysis.
• Technological progress: GPT-5’s success is likely to drive the development of new penetration testing platforms that combine AI with specialized tools.

However, the model’s high hacking effectiveness also raises concerns. Experts emphasize the need for strict oversight of such technologies to prevent their malicious use.

In Brief…

GPT-5 from OpenAI, tested on the XBOW platform, proved itself to be a “brilliant hacker,” detecting 70% of vulnerabilities in a single run and surpassing its predecessors in both speed and accuracy. Integration with a powerful platform revealed hidden capabilities related to improved logical reasoning and complex attacks. The research, conducted in August 2025, underscores GPT-5’s revolutionary potential in cybersecurity but also serves as a reminder of the importance of responsible AI use. This development opens a new era in security testing, promising more resilient systems—if the technology is applied wisely.