New Delhi, November 6, 2026: OpenAI has introduced GPT-6 Cyber, a specialised artificial intelligence model designed to identify software vulnerabilities, validate security weaknesses and help researchers develop fixes before cybercriminals can exploit them.
Announced on November 5, the model represents a major advancement in AI-powered cybersecurity. OpenAI has classified GPT-6 Cyber as having Critical cybersecurity capabilities under its Preparedness Framework, reflecting the model’s ability to discover and exploit sophisticated software vulnerabilities.
The company says the system has already helped researchers identify 81 previously unknown high-severity vulnerabilities in Mozilla Firefox and develop working exploits for 948 Linux kernel vulnerabilities.
What Is OpenAI’s GPT-6 Cyber?
GPT-6 Cyber is a specialised version of OpenAI’s GPT-6 family, developed for advanced cybersecurity research.
Unlike general-purpose AI chatbots, the model is designed to examine complex software systems, identify potential security weaknesses, reproduce vulnerabilities in authorised testing environments and assist in developing patches.
It can also perform extended security investigations involving large codebases and complicated technical systems.
OpenAI says the model has been specifically trained to remain within the boundaries of authorised security research.
GPT-6 Cyber Finds 81 Previously Unknown Firefox Vulnerabilities
One of the most significant demonstrations of the model’s capabilities involved Mozilla Firefox.
According to OpenAI, GPT-6 Cyber helped researchers discover 81 previously unknown vulnerabilities classified by Mozilla as high severity.
The model also identified 30 previously reported issues whose severity had been underestimated.
Researchers provided Mozilla with technical evidence, reproducible tests and proposed fixes.
Mozilla has already started releasing patches for many of the identified vulnerabilities.
The findings demonstrate how advanced AI systems could help security teams discover weaknesses in widely used software.
AI Model Develops Exploits for 948 Linux Vulnerabilities
GPT-6 Cyber was also tested against security weaknesses in the Linux kernel.
OpenAI reported that the model reproduced 1,724 vulnerabilities and successfully developed working exploits for 948 of them.
These findings included hundreds of previously unknown security issues.
The company said it had disclosed 304 critical findings to Linux maintainers, along with proposed patches.
The research demonstrates the model’s technical capabilities while also highlighting the importance of responsible vulnerability disclosure.
GPT-6 Cyber Outperforms Previous AI Models
OpenAI reported substantial performance improvements compared with its earlier GPT-5.6 Cyber model.
On ExploitGym, a cybersecurity benchmark measuring whether AI systems can exploit specified vulnerabilities, GPT-6 Cyber achieved a success rate of 48.5%.
GPT-5.6 Cyber scored 32.1% on the same benchmark.
That represents a relative improvement of approximately 51%.
On another evaluation involving web application vulnerability discovery, GPT-6 Cyber achieved 40.3% recall, compared with 32% for GPT-6 Astra.
These results indicate progress in the model’s ability to identify and validate complex cybersecurity weaknesses.
OpenAI Gives Model Critical Cybersecurity Rating
OpenAI has assigned GPT-6 Cyber a Critical cybersecurity capability classification under its Preparedness Framework.
The designation reflects the risks associated with highly capable AI systems that can discover vulnerabilities and develop working exploits.
However, the classification does not mean the model itself has been compromised or that it is being released without safeguards.
Instead, it indicates that its capabilities require stricter access controls, monitoring and misuse prevention.
OpenAI says it has introduced additional safety measures alongside the release.
Stronger Safeguards Aim to Prevent AI Misuse
One of the central concerns surrounding advanced cybersecurity AI is the possibility of misuse by malicious actors.
OpenAI says GPT-6 Cyber was trained to distinguish authorised testing from actions that exceed the permitted scope of a security investigation.
In one evaluation, the model avoided exploiting out-of-scope services in 99.93% of tests without additional deployment instructions.
With deployment safeguards enabled, it remained within the specified boundaries in 100% of the evaluated cases.
The company has also introduced monitoring systems designed to detect suspicious activity and prevent unauthorised use.
These results were obtained under controlled evaluation conditions and do not eliminate the possibility of misuse.
Access Restricted to Approved Cybersecurity Researchers
GPT-6 Cyber is not being made universally available.
The model is initially accessible through OpenAI’s API and Codex to selected customers approved for Daybreak Red, the company’s programme for advanced authorised cybersecurity research.
Access requires verification and additional organisational security controls.
OpenAI plans to expand availability to other approved Daybreak Red customers.
The company is also expanding access to earlier cybersecurity models and tools for security researchers.
Major Technology Companies Test GPT-6 Cyber
Several technology and cybersecurity companies have evaluated or started using the new model.
These include Cisco, Palo Alto Networks, SentinelOne, CodeRabbit and other security organisations.
The companies are exploring applications involving vulnerability discovery, security testing, code review and incident prevention.
OpenAI says the model can help researchers investigate complicated security problems that would otherwise require extensive manual analysis.
However, AI-generated findings still require appropriate validation before organisations make security decisions or deploy fixes.
OpenAI Focuses on Critical Infrastructure Protection
The company has also developed additional safeguards for sensitive infrastructure, including electricity networks, water treatment facilities and industrial control systems.
OpenAI collaborated with cybersecurity company Dragos to evaluate model behaviour in controlled industrial environments.
The evaluations examined how advanced AI systems might identify security weaknesses in operational technology.
OpenAI says it has strengthened controls designed to prevent assistance that could cause serious physical damage, disrupt essential services or endanger human life.
Approved critical infrastructure defenders may receive specialised access for authorised security testing.
AI Could Transform Cybersecurity Research
GPT-6 Cyber represents another step towards the use of AI systems capable of conducting increasingly complex security investigations.
Such technology could help developers identify weaknesses earlier, validate existing protections and prepare fixes more efficiently.
At the same time, the ability to discover and exploit vulnerabilities creates challenges for governments, software developers and security teams.
Experts have repeatedly warned that advanced AI capabilities could benefit both defensive researchers and malicious actors.
The impact of such systems will depend on responsible access, effective safeguards and cooperation with software maintainers.
What GPT-6 Cyber Means for the Future of AI Security
OpenAI’s latest release highlights the growing importance of artificial intelligence in cybersecurity.
The discovery of previously unknown vulnerabilities in Firefox and the successful reproduction of hundreds of Linux security flaws demonstrate the potential of specialised AI systems.
However, access restrictions and enhanced safeguards show that the technology also carries significant risks.
As AI becomes more capable of identifying complex software weaknesses, organisations may increasingly rely on automated security research to protect their digital systems.
The November 5 launch marks a significant development in the relationship between artificial intelligence, cybersecurity and responsible technology deployment.
Source: OpenAI Official Announcement, Reuters, CyberScoop, OpenAI GPT-6 Cyber System Card — November 5–6, 2026
Supreme News Network
