오픈AI 데이브레이크: 사이버 보안 특화 GPT-5.5-Cyber 공개

June 22, 2026 Security Company Daybreak: Tools for securing every organization in the world New tools, partnerships, and the full version of GPT‑5.5‑Cyber to move past vulnerability discovery and onto the acceleration of end-to-end patch automation. Loading… Share We’re expanding Daybreak ⁠ to help democratize patching vulnerable software at machine speed. For example, we’ve applied our models to discover and generate patches for critical vulnerabilities ⁠ in major browsers, network infrastructure, and operating systems such as FreeBSD and the Linux kernel. To scale the impact of these capabilities: Codex Security: We’re launching an update to the Codex Security plugin ⁠ , which implements what we’ve learned from internal and customer usage of our models into a solution to accelerate the process of discovering and patching vulnerabilities in existing systems as well as automatically preventing new vulnerabilities from ever reaching production. GPT‑5.5‑Cyber: Following an initial permissive-only preview, we’re launching the full version of GPT‑5.5‑Cyber through our continued limited release to trusted defenders. This model sets new state-of-the-art performance on CyberGym, reaching 85.6% compared with 81.8% for GPT‑5.5. Daybreak Cyber Partner Program ⁠ : Enabling security partners to scale the benefits to more organizations through our most capable models with trusted access in their products and services. Patch the Planet ⁠ : an initiative founded with Trail of Bits in collaboration with HackerOne, Calif, researchers, and maintainers to help widely used open-source projects move from findings to fixes. More than 30 open-source projects have committed to participate, with initial participants including cURL, Go, Python, Sigstore, and pyca/cryptography. With Patch the Planet, we are working with researchers, maintainers, enterprises, and partners to make powerful cyber capability available to defenders with appropriate access, governance, and human oversight. Hear from Clint and Dan about this here ⁠ (opens in a new window) . Cyber defense at an inflection point AI has changed the physics of cybersecurity. Frontier AI models have been increasingly accelerating vulnerability discovery. The bottleneck historically has been finding vulnerabilities, but now defenders are overwhelmed with the number of vulnerabilities found. Instead, the bottleneck is now patching vulnerabilities. For years, finding serious vulnerabilities required rare expertise, time, and deep familiarity with complex systems. Now, models can navigate large codebases, reason through attack paths, validate hypotheses, and surface security issues that might otherwise stay hidden. Defenders absolutely need access to these capabilities, and also need tools to fix what we can now find, before attackers do. Vulnerability reports, on their own, do not protect anyone. The value comes from validating the issue, understanding its impact, developing and testing a patch, coordinating disclosure, and helping teams deploy the fix. We are investing alongside our partners to improve these latter steps, in order to turbocharge defenders and convert model capability into real-world risk reduction. Frontier defensive capabilities should not be concentrated in the hands of a few. Software touches all aspects of life, from critical infrastructure to business applications and government networks. As AI changes the pace of vulnerability discovery, defenders everywhere need democratized access to these models to find, fix, and protect their infrastructure before attackers can identify and abuse these flaws. Daybreak brings together the frontier cyber capabilities OpenAI’s models, Trusted Access for Cyber, Codex Security workflows, and ecosystem partners to help approved defenders validate vulnerabilities, prioritize risk, generate and test fixes, and produce evidence inside existing security and development workflows. Our goal is to provide organizations the tools they need to stay secure even as the cyberthreat landscape continues to accelerate. From findings to fixes with Codex Security Since launching Codex Security cloud in research preview in March, it has scanned over 30 million commits across more than 30,000 codebases; human reviewers have manually marked more than 70,000 findings as fixed, and over 500,000 findings have automatically been determined to be fixed. This is the scale at which patching must now happen. We built Codex Security around a simple premise: put the equivalent of a security engineer next to every software developer by integrating directly into Codex. Rather than just generating alerts, Codex Security will understand your team’s code and its threat model (or generate one if it doesn’t exist), identify plausible vulnerabilities, determine whether affected code is reachable, gather evidence to provide validation steps, develop a targeted patch, and verify the result. Humans remain in control of which findings to investigate, which changes to apply, and what information to share. Today, we’re releasing an update to the Codex Security plugin ⁠ that enables out-of-the-box defensive security workflows. Developers can run deep scans or review recent changes, generate reports with severity, affected code locations, validation evidence, and remediation guidance, trace attack paths, build threat models, validate findings, and generate codebase-specific patches for review. Set up a scan to cover an entire codebase, a subset of the codebase, or a specific change or commit. The plugin ⁠ can also triage and validate existing findings from scanners, advisories, bug-bounty reports, or ticketing systems, then automate patch generation at scale to quickly close a backlog of vulnerabilities. When Codex Security completes a scan, it can also export to an existing vulnerability management system or integrate into tools with SARIF files, CodeQL queries, and more. The plugin makes these capabilities much more accessible to support automated pipelines with Codex CLI or integrate into developer workflows in the Codex app. Updating GPT‑5.5‑Cyber: pairing capability with permissiveness We are releasing an update to GPT‑5.5‑Cyber, our model that is both more permissive and more capable for advanced, authorized cybersecurity work. Our initial preview of GPT‑5.5‑Cyber was designed primarily to reduce unnecessary refusals in specialized workflows. This update goes further. It is our strongest model yet for finding and helping patch software vulnerabilities, while retaining GPT‑5.5’s general-purpose intelligence and ability to work across long, complex tasks. The model can sustain deeper analysis across large codebases: identifying security-relevant components, tracing whether vulnerable code is reachable, validating likely issues in controlled environments, developing and testing patches, and preparing evidence for human review. The goal is to help defenders move through the full remediation loop—not simply produce more findings. On CyberGym, which measures whether an agent can reproduce known vulnerabilities in software environments, the updated GPT‑5.5‑Cyber reached 85.6% in single-model evaluations, compared with 81.8% for GPT‑5.5. This is the highest CyberGym score we have measured from a single model. GPT‑5.5‑Cyber also outperformed GPT‑5.5 on two demanding real-world security benchmarks: 39.5% versus 25.95% on ExploitGym, which tests whether agents can turn known vulnerabilities into working exploits that achieve unauthorized code execution. On SEC-bench Pro, which evaluates long-horizon vulnerability discovery and proof-of-concept generation across complex software targets, GPT‑5.5‑Cyber reached 69.8%, compared with 63.1% for GPT‑5.5. Benchmarks are only one part of the story. What matters in practice is whether a model can find real vulnerabilities, distinguish actionable issues from noise, and help defenders land fixes safely. We are continuing to evaluate the model’s performance on complex repositories