Published: 11 August 2026
OpenAI launched GPT-5.6-Cyber on 10 August 2026, a cybersecurity-specific model built on GPT-5.6 Sol and available only through the new Daybreak Red access tier. GPT-5.6-Cyber is trained to reduce refusals on dual-use security work and completes 95.0% of requests on OpenAI’s internal Advanced Cybersecurity Completion Rate evaluation, against 1.5% for GPT-5.6 Sol. OpenAI used the model to find two previously unknown vulnerabilities in V8, the JavaScript engine in Chrome, which Google has patched as CVE-2026-15903.
What is GPT-5.6-Cyber and who can use it?
GPT-5.6-Cyber is a variant of GPT-5.6 Sol, OpenAI’s flagship model, trained specifically for cybersecurity workflows such as finding zero-day vulnerabilities and developing exploit chains. The point of the model is not only capability but permission: GPT-5.6 Sol refuses many highly dual-use security prompts, such as penetration testing against production systems, even when the system-level guardrails are lifted. GPT-5.6-Cyber is trained to answer those prompts for users OpenAI has vetted.
Access is restricted. The model is available only through Daybreak Red, and OpenAI controls entry through identity verification, account security requirements, monitoring, approved-use restrictions and legal attestations. At launch, OpenAI says GPT-5.6-Cyber went first to a group of trusted customer partners; press reporting names Accenture, IBM, CrowdStrike and Cloudflare among them, which OpenAI has not itself confirmed in the announcement. Organisations and individuals conducting authorised work can apply through OpenAI’s Daybreak partner programme. OpenAI has not published pricing for either tier.
Daybreak Blue and Daybreak Red: how the two access tiers differ
Daybreak is OpenAI’s cyber defence programme, launched earlier in 2026, and this release splits it into two tiers. Daybreak Blue gives approved defenders access to frontier general-purpose models including GPT-5.6 Sol, with the system-level safeguards that normally screen security prompts removed. OpenAI calls Blue the recommended starting point for most defenders and lists vulnerability discovery, secure code review, malware analysis, incident response and patch validation as the supported work.
Daybreak Red goes further. It grants access to OpenAI’s purpose-trained cybersecurity models for authorised vulnerability research, exploit validation and security testing, and it is the only route to GPT-5.6-Cyber. The distinction matters because the two tiers behave very differently on the same prompt: on OpenAI’s own completion-rate evaluation, GPT-5.6 Sol accessed through Daybreak Blue completes 2.0% of advanced requests, against 95.0% for GPT-5.6-Cyber through Daybreak Red. The predecessor model, GPT-5.5-Cyber, completed 57.3%, and OpenAI says the jump addresses researcher feedback about persistent refusals.
GPT-5.6-Cyber benchmarks and where it trails GPT-5.6 Sol
The 95.0% figure comes from Advanced Cybersecurity Completion Rate, an internal OpenAI evaluation that measures how often a model responds to requests involving exploit-chain development, authentication bypass, privilege escalation and similar scenarios. It measures willingness to answer, not accuracy. OpenAI has not published the prompt set or the sample size behind it, and no third party has reproduced it.
On capability, the results are mixed and OpenAI reports them without published score tables. On ExploitGym, which tests whether an agent can turn a known vulnerability into a working exploit that achieves arbitrary code execution, OpenAI says GPT-5.6-Cyber outperforms both GPT-5.6 Sol and GPT-5.5-Cyber. On an internal zero-day severity evaluation, GPT-5.6-Cyber also outperformed GPT-5.6 Sol accessed through Daybreak Blue. But on OpenAI’s Vulnerability Discovery and Report Writing evaluation, GPT-5.6-Cyber scores worse than GPT-5.6 Sol, which OpenAI attributes to the model producing shorter, less detailed vulnerability reports. On ExploitBench, a harder V8 exploitation task with the sandbox left enabled, GPT-5.6 Sol through Daybreak Blue performs best and more token-efficiently in the standard 300-turn setting, with the gap narrowing at 600 turns.
What GPT-5.6-Cyber found in Chrome’s V8 engine
Since the model finished training, OpenAI has used GPT-5.6-Cyber on real software. In V8, the JavaScript engine in Chrome, it found two previously unknown vulnerabilities that can be chained to corrupt memory and escape the V8 heap sandbox. OpenAI researchers validated the findings and reported them to Google under coordinated disclosure. Google has fixed the issue and assigned CVE-2026-15903, a high-severity bug in which the optimising compiler skipped a safety check when converting values to integers, so an undefined value could produce an unexpectedly large number. Used as an array index, that number could let the compiler omit a bounds check, allowing an attacker to read or overwrite memory belonging to other objects.
OpenAI also reports findings beyond Chrome, without naming the products: at least five vulnerabilities in a popular mobile operating system, including a chain from an untrusted app to local privilege escalation; three critical vulnerabilities in a popular database, including a remote path to code execution; and more than 400 privilege-escalation vulnerabilities in a popular operating system kernel. Those disclosures are being handled with Daybreak partners and open-source maintainers, so the claims cannot be independently checked yet.
Safety classification, controls and availability
Under OpenAI’s Preparedness Framework, GPT-5.6 Sol was assessed as High for cybersecurity capability and below the Critical threshold. OpenAI evaluated GPT-5.6-Cyber before launch and reached the same conclusion: High, but not Critical. That places it below Astra, the unreleased model OpenAI said last week may have reached the Critical cyber threshold. A system card with further evaluations of GPT-5.6-Cyber will be published at a later date, so the detailed safety results are not yet available.
OpenAI is pairing the launch with tighter operational controls. All individual Daybreak accounts must adopt hardware security keys from 1 September 2026. Daybreak customers using Codex are being pushed from full-access mode to auto-review mode, which evaluates actions needing elevated permissions before they run. OpenAI also recommends sandboxing security workflows away from production systems and the open internet, monitoring agent tool calls, and using scoped permission profiles. The release follows Anthropic’s cyber-focused model Mythos, which shipped earlier in 2026.
The full announcement, the evaluation detail and the access application can be found at the official OpenAI announcement on GPT-5.6-Cyber.
