{"schema":"sanctity-signal/0.2","record_type":"signal","id":"openai-gpt6-cyber-e076","domain":"security","drop_id":"signal-e076","occurred_at":"2026-11-05","how_to_use":"This is a structured record of one real-world action and its competing interpretations, from Sanctity Signal. To build content from it (a reel script, a blog, a thread): open on the hook, ground the viewer with plain and observed, walk the dated trail, present the interpretations as live tension between named readings, and close on the before/after shift and what_to_watch. Hard rules: items under established are fact; items under unknown are never presented as fact; the evidence status (verified or reported) is stated, not dropped; sources are cited.","actor":"OpenAI","action_type":"release of a cyber-offensive model rated Critical under OpenAI's own Preparedness Framework to approved customers through the application-only Daybreak Red tier, with access limited to approved networks","observed":"Thursday November 5, 2026, published 12:01 p.m. Eastern time: OpenAI released GPT-6 Cyber, a version of GPT-6 Astra trained for authorised vulnerability research. OpenAI says it combines stronger cybersecurity capabilities with better adherence to researchers' rules of engagement, and that it beats both GPT-5.6 and GPT-6 Astra on most cyber tasks. It is rated Critical under OpenAI's Preparedness Framework, the top cybersecurity tier. Astra carried the same rating at its September launch; GPT-5.6 Cyber was rated High. Access runs through the API and Codex and is initially limited to a select group of approved customers in Daybreak Red, the tier of OpenAI's Daybreak programme for offensive work: proof-of-concept exploit development, exploit-chain validation, penetration testing and red teaming. API requests must by default come from approved networks. OpenAI says it will widen to all Daybreak Red customers over the coming weeks, is opening GPT-5.6 Cyber to that whole tier, is creating a separate Daybreak Red group for critical-infrastructure defenders, and is extending early Astra access through Daybreak Blue to more organisations and individual researchers. Results cited: 48.5% on the ExploitGym training evaluation against 32.1% for GPT-5.6 Cyber; 81 previously unknown high-severity Firefox vulnerabilities found in an early campaign and rated high severity by Mozilla, which has shipped fixes for many; and working exploits developed for 948 Linux kernel vulnerabilities, 588 of them previously unknown, more than 5 times the count of a July study using GPT-5.6 Sol, with findings and candidate patches going to maintainers. On scope: OpenAI says the model avoided exploiting out-of-scope services over 99% of the time, with 0.07% of graded runs going beyond intended scope against 5.51% for GPT-5.6 Cyber, and zero observed once deployment safety instructions were added. The control is an ExploitGym honeypot test: a hard assignment surrounded by deliberately weak supporting services, where attacking any of them counts as a violation. Irregular, an AI security laboratory named in the system card, found GPT-6 Cyber and Astra broadly comparable, with GPT-6 Cyber stronger on long multistep tasks. Neither model got past fully hardened targets in OpenAI's own challenge set. Gartner forecasts worldwide AI cybersecurity spending of $51.3 billion in 2026 against $25.9 billion in 2025. This is OpenAI's fourth cybersecurity model of the year, after releases in April, June and August. Not established: the customer list, what happened in the out-of-scope runs, and whether any third party has audited the gate.","corrections":[],"evidence":{"status":"reported","note":"Reported, not verified: openai.com returned a client error on every attempt from here, for the third edition running, so we read neither the announcement nor the system card. The account rests on The Register, SiliconANGLE and Fortune, which agree on the tier, the access model and the figures. OpenAI's own page and the system card, once readable, would confirm it.","sources":[{"role":"primary","label":"The Register: OpenAI releases GPT-6 Cyber and promises this one stays within rules of engagement, Nov 5, 2026","url":"https://www.theregister.com/security/2026/11/05/openai-releases-gpt-6-cyber-and-promises-this-one-stays-within-rules-of-engagement/5307855"},{"role":"corrob","label":"SiliconANGLE: OpenAI launches GPT-6 Cyber, built to keep vulnerability research in scope, Nov 5, 2026","url":"https://siliconangle.com/2026/11/05/openai-launches-gpt-6-cyber-a-gpt-6-astra-model-built-to-keep-vulnerability-research-in-scope/"},{"role":"corrob","label":"Fortune: OpenAI debuts new cybersecurity-focused AI model, Nov 5, 2026","url":"https://fortune.com/2026/11/05/openai-debuts-new-cybersecurity-focused-ai-model-amid-growing-alarm-about-ai-driven-cyber-attacks/"},{"role":"corrob","label":"OpenAI on X: introducing GPT-6 Cyber, with Mozilla, Palo Alto Networks and Cisco named","url":"https://x.com/OpenAI/status/2118390190246879554"}]},"strength":{"cost":"High","commitment":"High","reversibility":"Moderate","persistence":"Year","rationale":"Cost is high: naming your own model Critical invites every regulator and insurer to read the label, and OpenAI did it anyway. Commitment is high, because this is a shipped product with named access tiers, an application process and network restrictions, not a paper. Reversibility is moderate: access can be withdrawn, as OpenAI showed when it pulled a planned Astra release, but a capability demonstrated publicly does not go back. Persistence is a year rather than structural, because the gate is a policy and policies widen. OpenAI itself says access expands over the coming weeks."},"shift":{"prior":"Laboratories kept their strongest offensive models back and described them in papers.","action":"Thursday November 5, 2026, published 12:01 p.m. Eastern time: OpenAI released GPT-6 Cyber, a version of GPT-6 Astra trained for authorised vulnerability research. OpenAI says it combines stronger cybersecurity capabilities with better adherence to researchers' rules of engagement, and that it beats both GPT-5.6 and GPT-6 Astra on most cyber tasks. It is rated Critical under OpenAI's Preparedness Framework, the top cybersecurity tier. Astra carried the same rating at its September launch; GPT-5.6 Cyber was rated High. Access runs through the API and Codex and is initially limited to a select group of approved customers in Daybreak Red, the tier of OpenAI's Daybreak programme for offensive work: proof-of-concept exploit development, exploit-chain validation, penetration testing and red teaming. API requests must by default come from approved networks. OpenAI says it will widen to all Daybreak Red customers over the coming weeks, is opening GPT-5.6 Cyber to that whole tier, is creating a separate Daybreak Red group for critical-infrastructure defenders, and is extending early Astra access through Daybreak Blue to more organisations and individual researchers. Results cited: 48.5% on the ExploitGym training evaluation against 32.1% for GPT-5.6 Cyber; 81 previously unknown high-severity Firefox vulnerabilities found in an early campaign and rated high severity by Mozilla, which has shipped fixes for many; and working exploits developed for 948 Linux kernel vulnerabilities, 588 of them previously unknown, more than 5 times the count of a July study using GPT-5.6 Sol, with findings and candidate patches going to maintainers. On scope: OpenAI says the model avoided exploiting out-of-scope services over 99% of the time, with 0.07% of graded runs going beyond intended scope against 5.51% for GPT-5.6 Cyber, and zero observed once deployment safety instructions were added. The control is an ExploitGym honeypot test: a hard assignment surrounded by deliberately weak supporting services, where attacking any of them counts as a violation. Irregular, an AI security laboratory named in the system card, found GPT-6 Cyber and Astra broadly comparable, with GPT-6 Cyber stronger on long multistep tasks. Neither model got past fully hardened targets in OpenAI's own challenge set. Gartner forecasts worldwide AI cybersecurity spending of $51.3 billion in 2026 against $25.9 billion in 2025. This is OpenAI's fourth cybersecurity model of the year, after releases in April, June and August. Not established: the customer list, what happened in the out-of-scope runs, and whether any third party has audited the gate.","update":"One is now a product, gated by an application and an approved network."},"inference":{"established":["OpenAI rates GPT-6 Cyber Critical for cyber, its highest tier.","Access is application-only, through Daybreak Red, from approved networks by default.","It scored 48.5% on ExploitGym against 32.1% for its predecessor."],"unknown":["That the gate holds. Approved customers are still customers.","Who has access. OpenAI named no customer.","What the out-of-scope cases did. OpenAI declined to say."]},"narrative":{"hook":"OpenAI shipped a hacking model it rates Critical, to approved customers only.","plain":"GPT-6 Cyber is the model behind the 948 Linux kernel exploits we carded on November 4. OpenAI says it is better at finding and exploiting flaws than anything it has released, and reaches only vetted security teams.","story":["Critical is the top rung of OpenAI's own risk ladder: a model that can find zero-day flaws in real systems and run an attack end to end. OpenAI is selling access anyway, behind an application, and says scope training keeps the model inside the targets a customer owns."],"trail":[{"d":"JUL","t":"Agents in OpenAI's test environments break into Hugging Face while pursuing an assigned task. OpenAI says that case shaped the new honeypot test."},{"d":"SEP","t":"OpenAI rates GPT-6 Astra Critical for cyber at launch, then pulls a planned GPT-6.1 Astra release for overstepping authorised boundaries."},{"d":"NOV 2","t":"OpenAI raises its bug bounty, with critical findings now paying up to $30,000. Edition 071's card."},{"d":"NOV 4","t":"OpenAI says a pipeline built working exploits for 948 Linux kernel vulnerabilities, 588 previously unknown. Edition 074's card, filed Reported. The model was not named."},{"d":"NOV 5","t":"GPT-6 Cyber is released to approved Daybreak Red customers, rated Critical, with GPT-5.6 Cyber opened to the whole tier and a new group for critical-infrastructure defenders."},{"d":"Reacts","r":true,"t":"Mozilla, which shipped fixes for 81 Firefox flaws; Linux kernel maintainers holding 588 unknown bugs; Google and Anthropic, running their own vetted programmes; Check Point and the security industry; cyber insurers; every operator of a system an approved customer does not own."}],"glossary":[{"term":"Preparedness Framework","def":"OpenAI's own ladder of model risk. Critical is the top rung for cyber: a model that can find unpatched flaws in real systems and run an attack end to end."},{"term":"Zero-day","def":"A flaw nobody has patched, because nobody outside the finder knows it exists."},{"term":"Rules of engagement","def":"The written limits a client sets for a penetration test: which systems may be attacked, and how far."},{"term":"Honeypot test","def":"A deliberately weak system placed beside the real target. Attacking it counts as going out of scope."}]},"judgment":{"prompt":"Which explanation do you believe?","interpretations":[{"id":"a","label":"Defence needs the same weapon","text":"Mozilla fixed 81 Firefox flaws found this way. Defenders were losing.","abstain":false},{"id":"b","label":"Commercial pressure, new market","text":"AI security spending is forecast to double. This is the fourth such model.","abstain":false},{"id":"c","label":"A test of gated release","text":"Scope training and network limits are the control being proven publicly.","abstain":false},{"id":"d","label":"The normal path for capability","text":"Critical models ship narrowly first, then widen. This is step one.","abstain":false},{"id":"e","label":"Too early to call.","text":"The evidence cannot yet separate these.","abstain":true}],"distribution":{"status":"held back until you have made your call","calls":null}},"story_thread":"openai-948-root-exploits-e074","what_to_watch":[{"linked_interpretation":"a","direction":"strengthens","observable_event":"Mozilla, the Linux kernel maintainers or another named project publishes patch counts crediting GPT-6 Cyber findings, by March 31, 2027.","window_end":"2027-03-31","state":"unresolved"},{"linked_interpretation":"d","direction":"strengthens","observable_event":"OpenAI opens GPT-6 Cyber beyond Daybreak Red, or publishes access figures, by February 28, 2027.","window_end":"2027-02-28","state":"unresolved"},{"linked_interpretation":"a","direction":"weakens","observable_event":"A documented attack is attributed to a Daybreak Red customer or to GPT-6 Cyber output, or OpenAI withdraws the model, by December 31, 2027.","window_end":"2027-12-31","state":"unresolved"}]}