← All drops
RESEARCHDrop 007 ·

OpenAI releases GPT-6 Cyber, and keeps it locked up

Watch this drop on TikTok

What happened

On 5 November 2026, OpenAI introduced GPT-6 Cyber. OpenAI says it is trained to find and check weak spots in software that attackers could use, while staying inside the limits that researchers set.

The Firefox result

OpenAI says researchers used the model to find 81 serious bugs in Firefox that nobody knew about before. Mozilla has released fixes for many of them. These are OpenAI's own figures.

Linux

OpenAI also says the model re-created known bugs in the Linux kernel. That means it reproduced problems that were already known, not that it found new ones.

The catch

GPT-6 Cyber is not going into regular ChatGPT. At launch, only a select group of approved Daybreak Red security teams can use it, through OpenAI's developer tools. OpenAI plans to widen this to all Daybreak Red customers.

Risk rating

OpenAI gives the model its highest risk rating for cybersecurity. That is one level above the rating of its earlier GPT-5.6 Cyber model.

  1. OpenAI: Introducing GPT-6 Cyber (5 Nov 2026)openai.com
  2. The Verge: OpenAI GPT-6 Cyber launchtheverge.com
  3. TechCrunch: OpenAI released a new cybersecurity model (5 Nov 2026)techcrunch.com