OpenAI has announced the release of GPT-6 Cyber, an AI model designed to help security researchers identify and address software vulnerabilities.
The company has classified the model as “Critical” in cybersecurity under its internal safety framework.
Access to GPT-6 Cyber is initially restricted to a group of vetted Daybreak Red customers through the API and Codex, with broader availability planned in the near future.
According to OpenAI, GPT-6 Cyber has been trained to discover and validate exploitable vulnerabilities while remaining within user-defined operational limits.
The model is intended to support security research by finding flaws, testing exploitation possibilities, and enabling the creation of patches before malicious actors can take advantage of weaknesses. This comes in response to the increased misuse of AI in cyberattacks.
In independent benchmarks, OpenAI reports that GPT-6 Cyber outperformed previous models.
On the ExploitGym test, which measures the ability to exploit specified vulnerabilities, GPT-6 Cyber achieved a score of 48.5%, compared to 32.1% for its predecessor, GPT-5.6 Cyber.
In alignment evaluations, the new model stayed within authorised task boundaries in 99.93% of tests without deployment prompts and 100% with such prompts.
OpenAI states that GPT-6 Cyber has already been applied to real-world software. The model was used to develop 948 privilege escalation exploits in Linux and found 81 previously unknown high-severity vulnerabilities in Mozilla Firefox.
The company disclosed these findings to Mozilla, which released patches for many of the issues.
The technology is also being utilised by security teams at organisations such as Calif, Cisco, and CodeRabbit for vulnerability research and code review.
OpenAI researchers used GPT-6 Cyber to reinforce their own infrastructure, including during red-teaming exercises that led to software patches and tighter controls.
OpenAI, in its announcement, said: “As models become more capable and uncover more security vulnerabilities, we are also seeing more attempts to use them to cause harm.”
The company has introduced strengthened safeguards alongside GPT-6 Cyber, including misalignment monitoring, IP allowlisting, and protections against high-risk cyber assistance.
The release of GPT-6 Cyber forms part of OpenAI’s broader Daybreak initiative, which provides security research tools and subsidises access for frontline defenders.
The company says its goal is to support security operations for organisations ranging from large enterprises to independent software maintainers.
Earlier this month, OpenAI lowered its target for its current funding round by $10bn. The move comes amid a wider downturn in technology stocks and growing investor concerns over spending on AI.
