--- title: OpenAI built a model it doesn’t want most people to use description: OpenAI released GPT-5.6 Cyber, a security-specialized model trained to perform tasks that general-purpose models refuse — including exploit chain development,... canonical: https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI built a model it doesn’t want most people to use | daily.dev og:description: OpenAI released GPT-5.6 Cyber, a security-specialized model trained to perform tasks that general-purpose models refuse — including exploit chain development,... og:url: https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly og:image: https://api.daily.dev/og/posts/f1L9eCnLy.png og:image:alt: OpenAI built a model it doesn’t want most people to use og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI built a model it doesn’t want most people to use **[The New Stack](https://daily.dev/sources/newstack)** · 4 min read · 0 upvotes · 0 comments ## Summary OpenAI released GPT-5.6 Cyber, a security-specialized model trained to perform tasks that general-purpose models refuse — including exploit chain development, zero-day hunting, and authentication bypass. It's available through Daybreak Red, a gated tier requiring identity verification, hardware security keys, and legal attestations. A lower tier, Daybreak Blue, gives approved defenders access to GPT-5.6 Sol with system-level restrictions removed for defensive tasks like malware analysis and patch validation. In internal testing, GPT-5.6 Cyber answered 95% of security requests versus 1.5–2% for Sol. The model found real vulnerabilities including two V8 flaws chained to escape Chrome's heap sandbox (CVE-2026-15903), plus critical flaws in a popular database and mobile OS. Early access partners include SpecterOps, SentinelOne, Palo Alto Networks, CrowdStrike, and Cisco. OpenAI rated the model 'High' on its Preparedness Framework cybersecurity scale, below the 'Critical' threshold. Hardware security key requirements for Daybreak users take effect September 1, 2026. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is OpenAI's Daybreak Red tier and what model does it provide access to? Daybreak Red is a gated cybersecurity program tier that provides access to GPT-5.6 Cyber, a model OpenAI trained specifically for advanced offensive security tasks including zero-day hunting, exploit chain development, and authentication bypass. In internal testing it answered 95% of security requests that GPT-5.6 Sol refused under both standard safeguards and the lower Daybreak Blue tier. Access requires identity verification, monitoring acceptance, legal attestations, and hardware security keys starting September 1, 2026. _Security teams evaluating AI-assisted red teaming tools track access requirements and capability benchmarks on daily.dev._ ### How does GPT-5.6 Cyber compare to GPT-5.6 Sol on security benchmarks? GPT-5.6 Cyber outperforms Sol on ExploitGym, which tests turning known flaws into working exploits in a sandbox, and answered 95% of exploit-related requests versus Sol's 1.5–2%. However, Sol outperforms Cyber on vulnerability writeup quality (Cyber's reports are shorter and less detailed), token efficiency on ExploitBench, and agent tasks capped at 300 turns. Cyber also uses a larger reasoning budget, making longer research sessions more expensive. _Developers choosing between OpenAI security tiers for their tooling find benchmark breakdowns like this on daily.dev._ ### What real vulnerabilities did OpenAI's GPT-5.6 Cyber find during testing? GPT-5.6 Cyber found two previously undisclosed flaws in V8, Chrome's JavaScript engine, that could be chained to corrupt memory and escape the V8 heap sandbox — patched by Google under CVE-2026-15903. It also found three critical vulnerabilities in a popular database including a remote code execution path, at least five in a popular mobile operating system, and over 400 possible privilege-escalation flaws in a popular OS kernel. Affected software names have not been disclosed as coordinated disclosure is ongoing. _Researchers tracking AI-discovered CVEs and coordinated disclosure timelines follow developments like these on daily.dev._ ## Similar posts on daily.dev - [OpenAI Previews GPT-5.6 Sol With Restricted Access and Stronger Cyber Safeguards](https://daily.dev/posts/openai-previews-gpt-5-6-sol-with-restricted-access-and-stronger-cyber-safeguards-kmzh7xyee) · Security Boulevard · 0 upvotes · 0 comments --- Tags: [#security](https://daily.dev/tags/security), [#cyber](https://daily.dev/tags/cyber), [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai), [#zero-day](https://daily.dev/tags/zero-day) [View this post on daily.dev](https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"OpenAI built a model it doesn’t want most people to use","url":"https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly"},"datePublished":"2026-08-10T21:21:01.900Z","dateModified":"2026-08-11T02:04:18.936Z","description":"OpenAI released GPT-5.6 Cyber, a security-specialized model trained to perform tasks that general-purpose models refuse — including exploit chain development,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4145c7f658eea4b595ccb7047f3c505e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4145c7f658eea4b595ccb7047f3c505e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,cyber,llm,openai,zero-day","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"OpenAI built a model it doesn’t want most people to use"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openai-built-a-model-it-doesn-t-want-most-people-to-use-f1l9ecnly#faq","mainEntity":[{"@type":"Question","name":"What is OpenAI's Daybreak Red tier and what model does it provide access to?","acceptedAnswer":{"@type":"Answer","text":"Daybreak Red is a gated cybersecurity program tier that provides access to GPT-5.6 Cyber, a model OpenAI trained specifically for advanced offensive security tasks including zero-day hunting, exploit chain development, and authentication bypass. In internal testing it answered 95% of security requests that GPT-5.6 Sol refused under both standard safeguards and the lower Daybreak Blue tier. Access requires identity verification, monitoring acceptance, legal attestations, and hardware security keys starting September 1, 2026. Security teams evaluating AI-assisted red teaming tools track access requirements and capability benchmarks on daily.dev."}},{"@type":"Question","name":"How does GPT-5.6 Cyber compare to GPT-5.6 Sol on security benchmarks?","acceptedAnswer":{"@type":"Answer","text":"GPT-5.6 Cyber outperforms Sol on ExploitGym, which tests turning known flaws into working exploits in a sandbox, and answered 95% of exploit-related requests versus Sol's 1.5–2%. However, Sol outperforms Cyber on vulnerability writeup quality (Cyber's reports are shorter and less detailed), token efficiency on ExploitBench, and agent tasks capped at 300 turns. Cyber also uses a larger reasoning budget, making longer research sessions more expensive. Developers choosing between OpenAI security tiers for their tooling find benchmark breakdowns like this on daily.dev."}},{"@type":"Question","name":"What real vulnerabilities did OpenAI's GPT-5.6 Cyber find during testing?","acceptedAnswer":{"@type":"Answer","text":"GPT-5.6 Cyber found two previously undisclosed flaws in V8, Chrome's JavaScript engine, that could be chained to corrupt memory and escape the V8 heap sandbox — patched by Google under CVE-2026-15903. It also found three critical vulnerabilities in a popular database including a remote code execution path, at least five in a popular mobile operating system, and over 400 possible privilege-escalation flaws in a popular OS kernel. Affected software names have not been disclosed as coordinated disclosure is ongoing. Researchers tracking AI-discovered CVEs and coordinated disclosure timelines follow developments like these on daily.dev."}}]} ```