{"rewrite":{"id":"r_7321f3b7a58286783f8bceb6","clusterId":"c_580fdf392753b13e2a969044","slug":"openai-ties-gpt-6-cyber-release-to-approved-daybreak-red-users","model":"deepseek-v4-1-flash","headline":"OpenAI Ties GPT-6 Cyber Release to Approved Daybreak Red Users","summary":"OpenAI has announced GPT-6 Cyber, a model it says performs 51% better than GPT-5.6 Cyber on the ExploitGym benchmark, which tests whether a model can exploit specified vulnerabilities in remote targets. The model found and verified nearly 1,000 exploits in Linux and Firefox, and avoided abusing external services not anticipated by humans in 99.93% of tests without safety prompts. It goes first to approved Daybreak Red users through the API and Codex.","whyItMatters":"OpenAI is shipping its strongest vulnerability-exploitation model behind an IP allowlist and an approved-user program, which makes access control, not benchmark score, the thing that decides who gets it first.","webCardHtml":"\u003cp\u003eGPT-6 Cyber found and verified nearly 1,000 exploits in Linux and Firefox, and OpenAI says it avoided abusing external services not anticipated by humans in 99.93% of tests run without safety prompts, up from 94.49% for GPT-5.6 Cyber. The model is provided through Daybreak Red, an advanced cybersecurity verification environment designed for security workflows such as developing exploits that work and attacking remote targets. Access needs an IP allowlist, and a feature called Private Safety Processing runs automated checks for fraudulent use in interactions with the AI. New user applications are open.\u003c/p\u003e","blueskyPost":"GPT-6 Cyber found nearly 1,000 verified exploits in Linux and Firefox, yet OpenAI limited the release to approved Daybreak Red users. The gated rollout is the real safety mechanism, not the benchmark score.","twitterPost":"GPT-6 Cyber's real safety test is the Daybreak Red access gate, not the 51% ExploitGym gain.","threadsPost":"GPT-6 Cyber's safety claim rests on that 99.93% figure, but the approved Daybreak Red access model is what OpenAI is actually relying on to keep 1,000 verified exploits from becoming a public tool.","newsletterBlurb":"OpenAI announced GPT-6 Cyber, a model it says found and verified nearly 1,000 exploits in Linux and Firefox and improved 51% on the ExploitGym benchmark over GPT-5.6 Cyber. Access runs through Daybreak Red, an advanced cybersecurity verification environment, starting with a limited group of approved users on the API and Codex. OpenAI is accepting applications for new Daybreak Red users.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20261106-openai-gpt-6-cyber/\",\"title\":\"OpenAI announces GPT-6 Cyber, an AI for cybersecurity\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":5498,"outputTokens":776,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1793937004,"createdAt":"2026-11-06T03:46:38.000Z","publishedAt":"2026-11-06T03:48:56.000Z","updatedAt":"2026-11-06T03:48:56.000Z"},"cluster":{"id":"c_580fdf392753b13e2a969044","canonicalTitle":"OpenAIがサイバーセキュリティのためのAI「GPT-6 Cyber」を発表","representativeArticleId":"a_3c55960436bc38083d532b28","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"announcement\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-11-06T02:59:00.000Z","lastSeenAt":"2026-11-06T02:59:00.000Z","updatedAt":"2026-11-06T03:48:52.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20261106-openai-gpt-6-cyber/","title":"OpenAIがサイバーセキュリティのためのAI「GPT-6 Cyber」を発表"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"announcement","domain":"other","is_roundup":false},"keyFacts":["OpenAI announced GPT-6 Cyber, a model it says performs 51% better than GPT-5.6 Cyber on the ExploitGym benchmark.","GPT-6 Cyber discovered and verified nearly 1,000 exploits in Linux and Firefox, according to OpenAI.","Without safety prompts, GPT-6 Cyber avoided abusing external services not anticipated by humans in 99.93% of tests, up from 94.49% for GPT-5.6 Cyber.","GPT-6 Cyber is provided through Daybreak Red, OpenAI's advanced cybersecurity verification environment, and Daybreak Red API access requires an IP allowlist.","GPT-6 Cyber goes first through the API and Codex to a limited group of approved Daybreak Red users."],"pageLinks":{"healthSlugs":[],"storySlugs":[],"entities":[],"stories":[],"entitiesOk":true,"storiesOk":true}}