OpenAI is offering a cybersecurity-focused model that TechCrunch calls “ChatGPT Cyber” to the top tier of its cybersecurity program, Daybreak Red, which gives offensive security researchers tools for finding and developing software vulnerabilities, according to TechCrunch, which reported on November 5, 2026. The OpenAI announcement the article links to has a URL that refers to “GPT-6-Cyber.”

Access

  • Daybreak Blue gives cyber defenders access to frontier models. Researcher Justicz used GPT 5.6 Sol, released in July, through it, and said he joined by uploading a photo of his driver’s license.
  • OpenAI says “thousands” of cybersecurity professionals take part, and that it wants to bring advanced security research to “organizations of every size,” along with independent researchers and open-source maintainers.
  • OpenAI announced a $1 billion initiative for program participants who protect critical infrastructure.

Bug-finding figures

  • TechCrunch says the same model was used on the Linux kernel, where it found 948 potential issues and helped prepare patches. The bug hunt cost $1.2 million, per the article, which compares that with roughly $6 million in tokens OpenAI spent on a fluid dynamics problem.
  • OpenAI reported 81 “novel” vulnerabilities in Firefox.
  • Justicz used the model to find a severe vulnerability that could let attackers take control of Ubuntu servers in about two hours, reported it to the Debian Project, and a patch was issued. TechCrunch says there is no known evidence of exploitation. Its CVE record is CVE-2026-109518, a different identifier from the one in Lens’s earlier post on an Ubuntu kernel flaw.
  • Justicz spent about $5,000 on subscriptions and tokens to find several exploits.

What the people quoted said

  • Justicz on the Ubuntu bug: “It’s extremely convoluted and hard for a human to find, but it works!” On what comes next: “We should expect some extremely crazy bugs to be found and exploited over the next few months.” He also said Daybreak Blue was recently “nerfed,” making it less useful for finding vulnerabilities.
  • OpenAI researcher Eric Wallace said the company is “fully pro bono” in devoting significant frontier compute to finding and patching open-source vulnerabilities.
  • OpenAI argues defenders have a narrow window to act before models get better at hacking. TechCrunch notes that defenders say restrictions hamper them while attackers can use open-weight models with safety restrictions removed, and the article questions whether frontier labs are devoting enough resources to the problem.

The article does not mention any cybersecurity model from Anthropic or Google.

Sources and notes

This post is based on TechCrunch’s report by Tim Fernholz (11:06 a.m. PST, November 5, 2026), which cites OpenAI’s announcements, the CVE record and Justicz’s social media thread. Lens did not read those sources, and the bug counts and costs are OpenAI’s and Justicz’s own figures as reported.

Image: open stainless-steel vault doors at the Tower of London, by Maggie Jones via Wikimedia Commons, released into the public domain by the photographer (Flickr Public Domain Mark); cropped. Stand-in photo; it is not related to OpenAI’s cybersecurity model and does not show the events described.