{"id":4018,"date":"2026-11-06T19:19:08","date_gmt":"2026-11-06T19:19:08","guid":{"rendered":"https:\/\/thedigitalfortress.us\/?p=4018"},"modified":"2026-11-06T19:19:08","modified_gmt":"2026-11-06T19:19:08","slug":"openai-launches-gpt-6-cyber-after-ai-helps-find-81-firefox-flaws-948-linux-exploits","status":"publish","type":"post","link":"https:\/\/thedigitalfortress.us\/?p=4018","title":{"rendered":"OpenAI Launches GPT-6 Cyber After AI Helps Find 81 Firefox Flaws, 948 Linux Exploits"},"content":{"rendered":"
\n

\ue804<\/i>Ravie Lakshmanan<\/span>\ue802<\/i>Nov 06, 2026<\/span><\/span>Artificial Intelligence \/ Software Security<\/span><\/p>\n<\/div>\n

\n
<\/a><\/div>\n

OpenAI on Thursday unveiled GPT\u20116 Cyber<\/strong>, which it described as a major step forward in security research that brings together stronger cybersecurity capabilities with a step-change improvement in model alignment.<\/p>\n

The new model is \u00abmore capable at finding and fixing vulnerabilities, and far more reliable staying within the boundaries researchers set,\u00bb OpenAI said<\/a>. It has classified the model as Critical in cybersecurity under its Preparedness Framework.<\/p>\n

The artificial intelligence (AI) company said GPT\u20116 Cyber<\/a> is also better than its predecessors like GPT\u20115.6 Cyber and GPT\u20116 Astra when it comes to following task boundaries and avoiding exploiting out-of-scope services in 99.93% of cases, up from 94.49% for GPT-5.6 Cyber.<\/p>\n

It also scores 48.5% on ExploitGym, a cybersecurity benchmark that evaluates the capabilities of AI agents to autonomously turn known software vulnerabilities into functioning, real-world exploits. In comparison, GPT-5.6 Cyber scored 32.1%.<\/p>\n

OpenAI said the model enabled the discovery of 948 local privilege escalation exploits in Linux\u2060 through an autonomous pipeline, along with 81 previously unknown high-severity vulnerabilities in Mozilla Firefox and a critical vulnerability in Brotli\u2060 (CVE-2026-111497, CVSS score: 9.2). <\/p>\n

GPT-6 Cyber is initially available through the API and Codex to select Daybreak Red<\/a> customers as part of efforts to make the best cybersecurity-focused models available with an aim to help secure software before bad actors have a chance to detect and exploit vulnerabilities using AI-assisted approaches. The risk is compounded by the growing abuse of open-weight models with abliterated guardrails.<\/p>\n

\u00abWhen tested without deployment safety prompts, it avoided exploiting out-of-scope services in 99.93% of cases, up from 94.49% for GPT\u20115.6 Cyber,\u00bb OpenAI said. \u00abWith those prompts, GPT\u20116 Cyber stayed within bounds 100% of the time.\u00bb<\/p>\n

\n
\"Cybersecurity\"<\/a><\/div>\n<\/div>\n

OpenAI said it has implemented additional guardrails to flag misuse and ensure the model’s actions fall within the scope of the user’s intent and authorization. As of last month, the company announced plans to use content from Daybreak Blue sessions on personal accounts starting November 29, 2026, to \u00abimprove our models, including to help strengthen safeguards against cyber misuse\u00bb even when the \u00abimprove the model for everyone\u00bb setting is off.<\/p>\n

<\/p>\n

\u00abTo prevent misuse of the Daybreak Red and Blue platforms, we have refined safeguards against assistance and capability development that could compromise industrial controls or safety-critical systems, interfere with equipment, or cause serious injury, catastrophic physical damage, or life-endangering service disruption,\u00bb it noted.<\/p>\n

\"\"<\/a><\/div>\n

OpenAI Reports New Misalignment Incidents<\/h3>\n

GPT-6 Cyber’s promise to stay within \u00abrules of engagement\u00bb comes as OpenAI disclosed<\/a> three misalignment incidents involving GPT-6 Astra and an internal research model. These incidents took place between July and October 2026 –<\/p>\n