OpenAI has begun offering GPT-6 Cyber, an AI model purpose-built to locate software vulnerabilities and develop exploits for them, to a small set of vetted customers. The release took place on Thursday.
The company has assigned the model a Critical rating for cybersecurity under its internal safety framework, the top level of concern in that category. The move, reported by The Economic Times, comes as OpenAI says attempts to misuse AI for malicious purposes are growing.
A Model Designed for Offensive Security Research
GPT-6 Cyber is intended to support security researchers through the full cycle of vulnerability work: finding weaknesses, testing how they could be exploited, and fixing them before attackers can take advantage.
That dual capability is what makes the model both useful and sensitive. Tools that can write working exploits help defenders understand real risk, but the same skills would be valuable to criminal groups.
Firefox and Linux Findings Show the Model's Reach
According to OpenAI, its researchers used the model to identify 81 previously unknown vulnerabilities in Firefox that Mozilla classified as high severity. Mozilla has started shipping fixes for them.
The model also helped produce working exploits for 948 Linux kernel vulnerabilities, 588 of which had not been known before. OpenAI said it has disclosed 304 critical Linux findings to the kernel's maintainers, along with suggested patches.
Why OpenAI Raised the Risk Rating
The Critical classification marks a step up from the High rating OpenAI gave its previous security-focused model, GPT-5.6 Cyber. The higher tier reflects the company's view that the new system's offensive capabilities carry greater potential for harm if misused.
OpenAI said it is observing more attempts to abuse AI, including by malicious cyber actors and by groups trying to distill its models. Distillation involves using a more capable model's outputs to train a cheaper copy, which could let restricted capabilities spread outside controlled channels.
Access Limited to the Daybreak Red Tier
GPT-6 Cyber is available only to selected customers of Daybreak Red, OpenAI's access tier for advanced and authorized security work such as penetration testing and exploit development.
The company plans to extend access to all approved Daybreak Red customers over the coming weeks. Starting Thursday, the earlier GPT-5.6 Cyber model is also available to every Daybreak Red customer.
What It Means for Defenders and Security Startups
The release signals that frontier AI labs now see automated vulnerability discovery and exploit development as mature enough to sell, but risky enough to gate behind vetting. Security teams with approved access may be able to find and patch bugs at a much larger scale than manual research allows.
For software maintainers, a surge of AI-generated vulnerability reports and proposed patches could become a new norm. For security startups, the gated-access model raises questions about how specialized tools will compete with, or build on top of, capabilities controlled by a few large providers.
Source: The Economic Times
