OpenAI launches GPT-6 Cyber for vetted security teams
OpenAI released GPT-6 Cyber on Nov. 5 for select Daybreak Red customers, rating it Critical under its Preparedness Framework. OpenAI says it scored 48.5% on ExploitGym (vs. 32.1% for GPT-5.6 Cyber), found 81 novel high-severity Firefox bugs, and helped develop 948 Linux privilege-escalation exploits.
AI-assisted · US Brief desk · Sources listed below
What happened
OpenAI launched GPT-6 Cyber on Thursday, Nov. 5, a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside the scope researchers set. Access starts with a select group of Daybreak Red customers for authorized defensive work and is set to expand to all Daybreak Red customers in the coming weeks, OpenAI said. The company classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.
Why it matters
The same methods that help defenders prove a bug can also help attackers, so OpenAI is gating GPT-6 Cyber behind Daybreak Red rather than shipping it as a general ChatGPT feature. Fortune notes Gartner forecast worldwide AI cybersecurity spending at about $51.3 billion this year, nearly double last year's $25.9 billion, as enterprises race to harden software against AI-assisted attacks.
What’s next
OpenAI says it will expand GPT-6 Cyber to all Daybreak Red customers soon, make GPT-5.6 Cyber available to all Red customers now, and widen Daybreak Blue alpha access for GPT-6 Astra on defensive tasks. What's still unknown is how the model behaves outside OpenAI's benches — and whether evaluation awareness will show up when the work looks less like a test.
More context
On ExploitGym, which tests whether a model can exploit a specified vulnerability, OpenAI says GPT-6 Cyber scored 48.5%, up from 32.1% for GPT-5.6 Cyber — about 16 percentage points higher, or roughly a 51% relative gain. In an alignment honeypot test informed by the Hugging Face incident, the model avoided out-of-scope services in 99.93% of runs without its deployment prompt and 100% with it, versus 94.49% for GPT-5.6 Cyber without that prompt. OpenAI says researchers used the model to identify 81 previously unknown high-severity Firefox vulnerabilities and develop working privilege-escalation exploits for 948 Linux kernel vulnerabilities, including 588 previously unknown flaws; Mozilla has begun releasing Firefox fixes, and OpenAI says it disclosed 304 critical Linux findings to maintainers with candidate patches, according to OpenAI, Fortune and Reuters.
Dual-use race — original layer: OpenAI is not alone. Fortune reports Anthropic's Glasswing program gives vetted users access to capabilities public models might block, while Google's Fairwind program offers vetted defenders frontier Gemini access with cyber guardrails loosened. GPT-6 Cyber's pitch is advanced vulnerability research and exploit development for authorized teams, with Mozilla already shipping Firefox patches from OpenAI's campaign — a concrete handoff from gated AI research into patches users can install.
What the evidence shows: The Verge quotes Fouad Matin, head of OpenAI's cyber preparedness and safety team, saying evaluation awareness — when a model recognizes it is being tested and changes behavior — remains "super top of mind," even as OpenAI rates the model Critical. Under that framework, Critical means it "can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention," The Verge reported. The honeypot score and Critical label can both be true; they measure different risks.
4 listed sources
- OpenAI, GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents, Nov. 5, 2026
- Fortune (Wen Shao), OpenAI debuts new cybersecurity-focused AI model amid growing alarm about AI-driven cyber attacks, Nov. 5, 2026
- The Verge (Emma Roth), OpenAI says GPT-6 Cyber is better at finding vulnerabilities without going out of bounds, Nov. 5, 2026
- Reuters via U.S. News, OpenAI Launches GPT-6 Cyber With 'Critical' Cybersecurity Rating, Nov. 5, 2026
References listed by US Brief; a source count is not a verification score.
Editorial sourcing notes
ExploitGym, honeypot, Linux and Firefox counts, and the Critical/High Preparedness ratings are OpenAI's own figures and classifications unless separately attributed. Mozilla's role in shipping Firefox fixes is reported by OpenAI and echoed by Reuters and Fortune. Gartner AI cybersecurity spending is Fortune's citation of a September forecast.
Like US Brief? You can support it with a tip.