OpenAI Launches Daybreak Red Team and Defense Factory to Test AI Cyber Security
By Ganesh Karki
Key facts
- Section
- Tech & AI
- Country
- USA
- Published
- 17 Nov 2026
OpenAI has launched Daybreak, a security research unit that uses AI to find and fix software vulnerabilities, under a new programme called Cyber at the Frontier. The announcement came on 12 November 2026.
Daybreak has two parts. The Red Team, now in early access, offers offensive security research and custom security assessments. The Defense Factory, a research preview, works on detection, triage and patching. Both are built around a new model, GPT-6 Cyber, which OpenAI says scored 48.5 percent on the ExploitGym benchmark, against 32.1 percent for GPT-5.6 Cyber.
OpenAI Daybreak at a glance
| Detail | Information |
|---|---|
| Announced | 12 November 2026 |
| Programme | Cyber at the Frontier |
| Red Team | Early access |
| Defense Factory | Research preview |
| Model | GPT-6 Cyber |
| ExploitGym score | 48.5% (GPT-5.6 Cyber: 32.1%) |
| Linux kernel bugs | 948 identified and validated, per OpenAI |
| Tool calls | 4,800x fewer than the next-best model, per OpenAI |
What happened
Daybreak has already been working with security partners and clients, and OpenAI says it has identified and validated 948 Linux kernel bugs. The company also says GPT-6 Cyber solved tasks with 4,800 times fewer tool calls than the next-best model.
The Red Team uses AI to attack a company's own systems, finding weaknesses before a real attacker does. The Defense Factory does the opposite work: spotting a weakness, sorting it by severity and preparing a patch, work that today is mostly done by people.
Why it matters
Cyber security is now at the centre of the AI arms race. Attackers are using AI to find weaknesses faster, and defenders need the same speed. OpenAI putting its strongest cyber model on both sides — attack testing and defence — shows how serious that race has become.
The launch also shows how AI companies are now selling security as a product, not just using it inside their own labs. OpenAI is turning its own red-team work into a service other companies can buy.
What is not confirmed
Daybreak has no published price, and it is not public who qualifies for early access. The 948-bug count and the benchmark figures are vendor-reported, not independently verified. GPT-6 Cyber is not generally available; it is in research preview and early access only.
Sources
The Register, 12 November 2026; AI Frontier Post; OpenAI, Cyber at the Frontier.
Sources and references
Information checked: 17 November 2026
Related stories
SpaceX Falcon 9 NROL-136 Launch: What to Watch on 21 November
SpaceX Falcon 9 NROL-136 launch is scheduled for 21 November from Vandenberg, carrying reconnaissance satellites for the National Reconnaissance Office.
20 November 2026AO3 Returns After 17 Days Offline After Security Breach
AO3 returns after 17 days offline following a security incident, with some features still disabled, and went down again briefly on 19 November for maintenance.
20 November 2026ChatGPT for Accounting Launches With QuickBooks Link
ChatGPT for Accounting launched this week, connecting the chatbot to QuickBooks, Xero and NetSuite so finance teams can reconcile books and close the month.
20 November 2026Tesla Cybercab Rides Begin in Dallas With V1.5 Cars
Tesla Cybercab rides began in Dallas on 19 November, the second city after Austin to get the steering wheel-free robotaxi, with upgraded V1.5 vehicles.