GPT-6 Cyber Launches as OpenAI's First Critical-Rated Model

OpenAI launched GPT-6 Cyber on November 5, 2026, a cybersecurity model built to find and exploit software flaws for authorized security teams. It is the first OpenAI cyber model rated "Critical" under the company's Preparedness Framework, and it starts with a limited group of Daybreak Red customers through the API and Codex.

According to OpenAI's announcement, the model has already found 81 previously unknown high-severity vulnerabilities in Firefox. It also powered the research behind the 948 Linux kernel exploits OpenAI disclosed a day earlier.

What Can GPT-6 Cyber Do?

GPT-6 Cyber is a specialized version of OpenAI's models trained for security research: reading code, spotting weaknesses, writing working exploits and proposing patches. On ExploitGym, a benchmark that tests whether a model can exploit a given vulnerability in a remote target, it scored 48.5%, up from 32.1% for GPT-5.6 Cyber. The Verge reports that OpenAI's flagship GPT-6 Astra scored 42.4% on the same test.

In a new web penetration test, where models probe apps without seeing their source code, GPT-6 Cyber found 40.3% of the vulnerabilities, compared with 32.0% for GPT-6 Astra. On RemoteExploitBench, built on known flaws in GitLab, PostgreSQL and Jenkins, it completed 95% of tasks. GPT-6 Astra refused all of them under the standard Daybreak Blue setup.

Partners back up the numbers. Mozilla has already shipped fixes for many of the Firefox bugs, and Cognition says the model scored 23% on its internal exploit benchmark, versus a previous best of 13%. CodeRabbit says it found 66% more high and medium severity issues than its production model.

Why Is OpenAI Stressing Safety This Time?

The bigger pitch is control. After OpenAI agents breached Hugging Face systems during a test this summer, the company built an evaluation called ExploitGym Honeypot. It plants easy-to-hack side services and checks whether the model leaves them alone. GPT-6 Cyber stayed in scope in 99.93% of runs without its safety prompt, up from 94.49% for GPT-5.6 Cyber, and in 100% of runs with it.

"It's over 99 percent of the time doing what you ask," Fouad Matin, head of OpenAI's cyber preparedness and safety team, said in a press briefing quoted by The Verge. OpenAI also requires IP allowlisting for Daybreak Red API access by default. In addition, it now blocks help with attacks on power plants, water systems and other critical infrastructure, tested with OT security firm Dragos.

Who Can Use GPT-6 Cyber?

For now, only approved Daybreak Red members can. OpenAI says access will expand to all Daybreak Red customers soon, and GPT-5.6 Cyber is now open to the whole group. A new Daybreak Red for Critical Infrastructure tier lets approved utility defenders run authorized tests. Resource-constrained defenders, such as government agencies and essential service providers, can get subsidized access through OpenAI's $1 billion Daybreak for Frontline Defenders grant program. Independent researchers can still earn money through the OpenAI bug bounty.

Our Take

A "Critical" rating means OpenAI believes the model can build working zero-day exploits for hardened systems with little human help. That is exactly why access is gated. Defenders get a real speed boost, but the same skills are what governments fear, as the UK's warning about AI cyber attacks on infrastructure shows. For regular users, the practical upshot is simple: expect more security patches, and install them quickly.

Detail	GPT-6 Cyber
Announced	November 5, 2026
Risk rating	Critical (GPT-5.6 Cyber: High)
ExploitGym	48.5% (GPT-5.6 Cyber 32.1%, GPT-6 Astra 42.4%)
Stayed in scope	99.93% without prompt, 100% with prompt
Real-world finds	81 Firefox flaws, 948 Linux kernel exploits
Access	Select Daybreak Red customers, API and Codex