The specialized model also helped develop hundreds of Linux exploits, highlighting how faster vulnerability discovery could reshape cybersecurity research and increase pressure on software maintainers.
OpenAI has introduced GPT-6 Cyber, a specialized artificial intelligence model designed to identify, validate and help remediate software vulnerabilities, as concerns grow over the increasing capabilities of AI-powered cybersecurity systems.
The November 5 launch disclosed that OpenAI researchers used the model to identify 81 previously unknown high-severity vulnerabilities in Mozilla Firefox and develop working exploits for 948 Linux kernel vulnerabilities.
OpenAI classified the model’s cybersecurity capabilities as Critical, its highest designation under the company’s Preparedness Framework.
The release reflects a significant development for software engineers: artificial intelligence can now automate portions of vulnerability research that previously required sustained specialist investigation.
Firefox and Linux Reveal the Model’s Capabilities
According to Reuters, the Linux testing included 588 vulnerabilities that researchers described as previously unknown.
OpenAI also disclosed 304 critical Linux findings to maintainers, together with proposed patches.
The company’s research involved reproducing vulnerabilities and demonstrating how they could potentially be exploited within controlled environments.
Meanwhile, Mozilla’s Firefox 159 security advisory documents numerous high-severity vulnerabilities reported by OpenAI’s security researchers, including memory-safety and JavaScript engine issues.
Mozilla has released fixes for many of the reported Firefox vulnerabilities.
However, the results do not mean every identified Linux vulnerability has been remediated or that attackers have exploited them in real-world systems.
Cybersecurity Models Are Becoming More Capable
OpenAI reported that GPT-6 Cyber achieved a 48.5% success rate on ExploitGym, compared with 32.1% for GPT-5.6 Cyber.
That represents an improvement of approximately 16 percentage points.
The model also performed better in OpenAI’s evaluations of whether AI systems respected the boundaries of authorized security tasks.
In one internal evaluation, it avoided targeting out-of-scope services in 99.93% of tests without a deployment prompt.
These are company-reported benchmark results and should not be interpreted as guarantees of safe behavior in every operational environment.
Advanced Cybersecurity Research Comes With Higher Costs
Unlike general-purpose AI tools, GPT-6 Cyber is initially available only to selected customers through OpenAI’s Daybreak Red program for approved security research.
According to VentureBeat’s technical coverage, API pricing starts at $20 per million input tokens and $100 per million output tokens.
The publication noted that the model’s pricing exceeds competing specialized cybersecurity offerings from Anthropic and Google.
OpenAI has also described an automated Linux vulnerability research pipeline that produced 948 working privilege-escalation exploits at an API cost of approximately $1.2 million.
The figure illustrates both the scale of automated security testing and the considerable computing expense involved.
Faster Discovery Creates a New Security Problem
For developers, the most important implication is that vulnerability discovery may increasingly outpace the ability of organizations to review findings, prepare fixes and deploy updates.
Security teams must distinguish genuine, exploitable vulnerabilities from less consequential findings while avoiding disruptions caused by rushed patches.
For software maintainers, AI-generated reports could create additional workloads unless accompanied by reproducible evidence and reliable remediation proposals.
The technology also raises concerns about misuse if comparable capabilities become available to malicious actors.
The emerging cybersecurity challenge is therefore not simply finding vulnerabilities faster. It is ensuring defenders can verify, prioritize and repair those weaknesses before attackers exploit them.
