OpenAI Puts GPT-6 Cyber in the Hands of Vetted Daybreak Red Teams
OpenAI releases GPT-6 Cyber to select Daybreak Red security teams with no public pricing, citing 81 Firefox flaws and 948 Linux kernel exploits.
Listen
OpenAI has begun releasing GPT-6 Cyber, a security research model, to select Daybreak Red customers, with access set to widen over time. The release came on November 5, 2026, and the company has not published public pricing. Security teams at Mozilla, Palo Alto Networks and Cisco are among the named users.
GPT-6 Cyber is built to identify, validate and help patch software vulnerabilities. According to OpenAI's claims, it found 81 new high-severity vulnerabilities in Firefox and developed exploits for 948 Linux kernel issues. That Linux work underpins research OpenAI announced the previous day.
Access runs through the API and Codex and is limited to approved Daybreak Red members. OpenAI is subsidizing access for frontline defenders through a $1 billion grant program. One third-party write-up cites token rates of $20 and $100 per million, but those figures could not be confirmed against OpenAI materials, so the company's pricing position remains unresolved.
How GPT-6 Cyber Compares
On ExploitGym, OpenAI reports a score of 48.5% for GPT-6 Cyber, up from 32.1% for its predecessor, GPT-5.6 Cyber. In black-box web penetration tests, the model caught 40.3% of vulnerabilities, and it finished 95% of the jobs in RemoteExploitBench, a set covering software such as GitLab and Jenkins. OpenAI also says the model outperforms GPT-5.6 Cyber and GPT-6 Astra on most cybersecurity tasks. The model is classed as Critical under OpenAI's Preparedness Framework.
The sources reviewed here name no rival security model in direct comparison, so the competitive picture rests on OpenAI's own internal benchmarks against its earlier releases.
Safety Controls and Their Cost
Training included reinforcement learning aimed at preventing reward hacking and unauthorized sandbox escapes. New measures include identity verification, periodic check-ins for high-risk requests and enhanced monitoring. Customers can opt out of having their Daybreak usage data used for training.
For security teams, these controls add onboarding and review steps before a model with exploit-writing capability can be used. That makes vetting a cost in its own right, alongside any eventual price.
The benchmark figures are OpenAI's own claims, and the sources reviewed include no independent third-party evaluation. The sources also do not say how many of the 948 Linux kernel exploits concern previously unknown flaws, and OpenAI has not given a date for broader access.
CISOs evaluating the model should treat the ExploitGym and recall figures as vendor claims and watch for the first published price and the next expansion of access. Until then, the $1 billion grant program is the only disclosed route to lower costs for frontline defenders.
Sources
GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents
Photo by Brecht Corbeel on Unsplash
Related Articles
- GPT-5.6-Cyber: Reduced-Refusal Power Brings New Risk
- OpenAI Daybreak Launches with GPT-5.5-Cyber to Automate Enterprise Defense
- GPT-6 Cyber Preview Nears as OpenAI Prepares Its First Security Deployment Tool
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.