AI & Emerging Tech • • 3 min read

New GPT-6 Cyber ​​finds security holes faster than any hacker

OpenAI announced GPT-6 Cyber, an artificial intelligence model designed for advanced cybersecurity research. The tool was trained to identify vulnerabilities, test forms of exploitation and ...

Noah Patel
Noah Patel
Senior How-To, Troubleshooting & Tutorials Editor
New GPT-6 Cyber ​​finds security holes faster than any hacker
New GPT-6 Cyber ​​finds security holes faster than any hacker — Editorial Visual / Silicon Dispatch US
⚡ KEY DISPATCH • BOTTOM LINE UP FRONT (BLUF)

OpenAI announced GPT-6 Cyber, an artificial intelligence model designed for advanced cybersecurity research. The tool was trained to identify vulnerabilities, test forms of exploitation and suggest code fixes. According to the company, the model outperforms GPT-5.6 Cyber ​​and can find critical flaws before attackers have the opportunity to exploit them.

FTC Affiliate Disclosure: Silicon Dispatch is reader-supported. When you buy through verified retail links on our site, we may earn an affiliate commission at no extra cost to you. Prices and promo codes are verified at publication time.
US Verified Retailer
Check Live Price
Get Deal at Store ↗

BLUF (Bottom Line Up Front): OpenAI announced GPT-6 Cyber, an artificial intelligence model designed for advanced cybersecurity research. The tool was trained to identify vulnerabilities, test forms of exploitation and suggest code fixes. According to the company, the model outperforms GPT-5.6 Cyber ​​and can find critical flaws before attackers have the opportunity to exploit them.

During security testing, GPT-6 Cyber ​​identified 81 previously unreleased vulnerabilities in the Firefox browser, all classified by Mozilla as high severity. The model also found that another 30 known problems were more dangerous than initial assessments indicated. The company has already released fixes for many of these flaws.

In the Linux ecosystem, AI reproduced 1,724 vulnerabilities and developed codes (exploits) capable of exploiting them in 948 cases. The result surpasses the 166 exploits obtained by GPT-5.6 Sol in the same testing environment. OpenAI also forwarded 304 critical findings to Linux kernel maintainers.

The company even tested the model on its own infrastructure. During controlled security exercises, actors with access to GPT-6 Cyber ​​were able to patch vulnerabilities, escape isolated environments (microVMs), and execute code on the underlying servers. The tests helped OpenAI identify problems and reinforce infrastructure protection mechanisms.

In the ExploitGym benchmark, which measures the ability to exploit security flaws, GPT-6 Cyber ​​achieved a 48.5% score, compared to 32.1% for GPT-5.6 Cyber. In web application penetration tests, the model identified and exploited 40.3% of the vulnerabilities assessed.

OpenAI also evaluated whether the AI ​​respects the limits of authorized activities. In the ExploitGym Honeypot test, GPT-6 Cyber ​​avoided attacking services that were outside the defined scope in 99.93% of attempts, even when it was not given specific deployment instructions. When these guidelines were active, the model respected the limits in 100% of cases.

The increase in exploration capabilities led OpenAI to raise the model’s risk rating from “High” to “Critical”. The category considers the potential for using the technology in activities that may pose security risks.

GPT-6 Cyber ​​is not available to the general public. Due to risk classification, OpenAI has restricted access to verified security teams, authorized professionals, and partner organizations.

Silicon Dispatch Strategic Takeaways

  • Desk Alignment: Verified dispatch within the AI & Emerging Tech pillar.
  • Technical Verification: Hardware and software specifications corroborated against manufacturer roadmaps.
  • Market Impact: Analysis tailored for North American enterprise decision-makers and high-performance tech enthusiasts.

Affiliate Disclosure: Articles published by Silicon Dispatch may include verified affiliate partner links. If you purchase through these links, we may earn an affiliate commission at no additional cost to you.

Read Next on Silicon Dispatch

View all stories →