# OpenAI Releases GPT-5.5-Cyber in Daybreak Expansion for Cyber Patching

> The June 22, 2026 update features the full model release, Codex Security enhancements, and a new initiative to secure open source projects through AI-driven patching under controlled access.

*Published 2026-06-22 · By Marcus Vance*

GPT-5.5-Cyber is OpenAI's frontier model specialized in cybersecurity that supports machine-speed patching of open-source software vulnerabilities.

OpenAI announced the expansion of its Daybreak platform on June 22, 2026. This announcement included an update to the Codex Security plugin, the full release of the GPT-5.5-Cyber model, and the launch of the Patch the Planet initiative in partnership with Trail of Bits. The development represents a strategic move to apply frontier AI models to the challenge of securing open source software. By transitioning from vulnerability discovery to active patching at machine speed, the platform addresses a critical need in the cybersecurity landscape. The full release of GPT-5.5-Cyber under trusted access controls ensures that the most capable version is used responsibly. This approach allows for greater impact while managing potential risks associated with advanced AI capabilities in security contexts. The integration of these elements is expected to accelerate the remediation of vulnerabilities across a wide range of projects.

The background for this expansion lies in the increasing reliance on open source software across industries. Many organizations use these components without the resources to maintain their security independently. AI tools offer a way to augment those efforts by providing automated analysis and suggestion capabilities. The Daybreak platform was initially introduced to provide such tools, and the current expansion builds on early feedback and results from the research preview phase. The focus on patching rather than just detection marks an evolution in how AI is deployed for defensive purposes. This shift is timely given the pace at which new vulnerabilities are discovered and the potential damage they can cause if left unaddressed.

Cybersecurity threats continue to evolve, with attackers leveraging automated tools themselves to find and exploit weaknesses in code. Defenders need corresponding advancements to keep up. The Daybreak initiative seeks to provide that edge by harnessing the pattern recognition and generation abilities of large language models. The partnership with Trail of Bits brings practical experience in security auditing to the AI-driven process. This combination is intended to produce actionable results that maintainers can implement quickly. The overall goal is to strengthen the foundational software that supports modern technology infrastructure.

## What is the background and purpose of the Daybreak platform?

The Daybreak platform serves as a hub for AI-powered security tools aimed at organizations of varying sizes. Its purpose is to democratize advanced capabilities that were previously available only to well-resourced entities. By expanding the platform, OpenAI is responding to the demand for scalable solutions to software security challenges. The platform's tools are designed to integrate into existing workflows, allowing security teams to leverage AI without overhauling their processes. This accessibility is crucial for addressing the long tail of open source projects that receive limited attention from security researchers. The initiative also highlights the potential for public-private partnerships in enhancing digital security.

Purposefully, the platform focuses on practical outcomes such as reduced exposure times and improved code quality. The expansion incorporates lessons from the research preview, where user feedback informed the decision to release the full model. The emphasis on trusted access reflects an understanding of the dual-use nature of powerful AI models. Users must meet certain criteria to access the most advanced features, ensuring that the technology is used for defensive rather than offensive purposes. This framework is part of a broader strategy to promote responsible AI development in sensitive domains like cybersecurity.

## What specific components were introduced in the June 22 announcement?

The June 22, 2026 announcement detailed three main components. First, an update to the Codex Security plugin enhances its scanning and analysis functions. Second, the full GPT-5.5-Cyber model provides improved performance in vulnerability handling tasks. Third, the Patch the Planet initiative establishes a structured program for applying these tools to real open source projects. Together, these elements create a comprehensive offering for users interested in AI-assisted security. The announcement also provided details on the benchmark results and the initial outcomes from the partnership with Trail of Bits. This transparency allows the community to assess the effectiveness of the new releases.

The update to the Codex Security plugin builds on its previous capabilities by incorporating insights from the newer model. This allows for more accurate identification of potential issues in code commits. The full release of GPT-5.5-Cyber means that users with appropriate access can now utilize its complete feature set for tasks such as generating patch suggestions. The Patch the Planet initiative operationalizes these capabilities by facilitating direct engagement with project maintainers. The initial sprint demonstrated the feasibility of this approach through concrete deliverables like pull requests and merged changes. These components are interconnected, with the model and plugin supporting the initiative's goals.

## What performance metrics were reported for GPT-5.5-Cyber?

Performance on the CyberGym benchmark is a central metric for evaluating GPT-5.5-Cyber. The benchmark is structured around 1,507 tasks that simulate the reproduction of known vulnerabilities in 188 open source projects. Success in these tasks requires the model to navigate complex codebases and identify the conditions under which vulnerabilities manifest. The 85.6% score indicates a high level of proficiency across this diverse set of challenges. This result surpasses previous model performances and sets a benchmark for future developments in AI for security. The evaluation methodology ensures that the score reflects practical utility rather than theoretical capabilities.

The benchmark's design emphasizes real-world conditions, including variations in project structure and programming languages. Models must demonstrate the ability to handle tasks that mirror the work of human security researchers. Achieving the top score in this environment suggests that GPT-5.5-Cyber can contribute meaningfully to vulnerability management processes. The improvement from earlier versions, such as the 81.8% mentioned in comparisons, shows the iterative progress in model training and fine-tuning for cyber tasks. This progress is important for building confidence in AI tools among practitioners.

## What results emerged from the Patch the Planet initiative?

The Patch the Planet initiative's initial sprint focused on 19 open source projects with the support of Trail of Bits engineers. The application of GPT-5.5-Cyber and Codex Security led to the identification of hundreds of security issues. This effort resulted in 64 pull requests being submitted to address the identified problems. Additionally, 51 issues were filed to bring attention to other concerns that may require further investigation. The fact that 37 of the patches have already been merged demonstrates the practical value of the AI-generated suggestions. These outcomes were achieved in the first week of the initiative, indicating rapid progress.

Beyond the immediate fixes, the initiative also produced reusable workflows that can benefit other projects. Examples include fuzzing labs for automated testing and variant analysis pipelines for exploring related vulnerabilities. These workflows leverage the strengths of the AI models to create efficient processes that security teams can adopt. The collaboration between OpenAI and Trail of Bits exemplifies how partnerships can amplify the impact of frontier models. The results provide a template for scaling such efforts to additional projects in subsequent phases of the initiative.

Key Metrics from the Daybreak Expansion and Patch the Planet InitiativeMetricValueAssociated EntityCyberGym Benchmark Score85.6%GPT-5.5-CyberCommits Scanned by Codex SecurityOver 30 millionCodex SecurityCodebases AnalyzedMore than 30,000Codex SecurityPull Requests Generated64Patch the PlanetIssues Filed51Patch the PlanetPatches Merged37Patch the Planet

- Dedicated security engineers from Trail of Bits collaborated with OpenAI tools on 19 projects.
- Hundreds of security issues were identified through AI-assisted analysis.
- 64 pull requests were submitted to address the discovered vulnerabilities.
- 51 issues were filed to document additional concerns for maintainers.
- 37 patches were successfully merged into the open-source codebases.
- Reusable workflows such as fuzzing labs and variant analysis pipelines were developed for ongoing use.

## What operational scale has the Codex Security plugin reached?

The Codex Security plugin has operated at a significant scale since entering its research preview phase. It has scanned over 30 million commits across more than 30,000 codebases. This extensive coverage allows for the detection of issues that might otherwise go unnoticed in the vast landscape of open source development. The processing of these scans has led to more than 70,000 findings being marked as fixed by human reviewers. Such numbers highlight the plugin's role in augmenting human capabilities in security review processes. The data generated from these operations also contributes to the ongoing improvement of the AI models.

The scale of operation is indicative of the plugin's integration into development pipelines at various organizations. By analyzing historical commits, the tool can identify patterns of vulnerabilities that have been introduced over time. This retrospective analysis complements forward-looking security measures. The volume of findings processed underscores the need for efficient tools that can handle large amounts of data without overwhelming human analysts. The plugin's performance in this regard supports its continued use and expansion within the Daybreak platform.

## What are the implications for the market and stakeholders?

The release has significant implications for the cybersecurity market. Open source maintainers gain access to advanced tools that can supplement their limited resources. This can lead to more secure software being available for widespread use. Enterprises that build upon open source components benefit from reduced risk of supply chain attacks stemming from unpatched vulnerabilities. The market may see increased investment in AI security tools as the demonstrated results encourage adoption. However, the trusted access model may limit immediate availability to certain users, affecting the pace of broader market penetration.

Stakeholders including regulators and standards bodies may look to these developments for guidance on AI use in security. The balance between capability and control could inform policy discussions around AI deployment. For developers, the availability of patch suggestions from AI could change how code reviews are conducted. The overall effect is likely to be a more dynamic security environment where AI plays a larger role in maintaining software integrity. This evolution requires careful management to ensure that the benefits are realized without introducing new vulnerabilities through the AI systems themselves.

## What expert reactions have been shared about the release?

Expert reactions have focused on the potential for these tools to transform security operations. The ability to move at machine speed on patching tasks is viewed as a major advantage in a field where timing is critical. At the same time, experts caution that the integration of frontier models must be accompanied by appropriate safeguards. The comments from industry leaders reflect an awareness of both the opportunities and the challenges presented by this technology. This measured response is typical in the adoption of new security technologies.

> At Cisco, we view frontier models as a powerful force multiplier for defenders. Models like GPT-5.5 are fundamentally changing the velocity of our operations, enabling us to move faster on everything from incident investigation to proactive exposure reduction. But speed cannot be traded for trust.Anthony Grieco, SVP, Chief Security & Trust Officer, Cisco

The pullquote from Anthony Grieco captures the sentiment that while AI can enhance velocity, trust remains non-negotiable. This perspective is shared by many in the security community who have experience with previous waves of technology adoption. The reactions suggest that the release is seen as a positive step, provided that the controls around access are maintained and effective. Continued dialogue between model developers and end users will be important for refining these tools based on practical experience.

## What is anticipated for the future of AI in cybersecurity?

Looking ahead, the Daybreak platform is expected to evolve with additional features and partnerships. The success of the initial Patch the Planet sprint may lead to expansion to more projects and potentially different types of software. Refinements to the GPT-5.5-Cyber model based on feedback from these operations could yield further performance gains on benchmarks like CyberGym. The development of more specialized plugins and workflows is also likely as the ecosystem matures.

The field of AI for cybersecurity is poised for continued growth as more organizations recognize the value of these tools. The current release provides a foundation for exploring new applications, such as predictive security measures or integration with other defensive systems. As the technology advances, the emphasis on responsible use and trusted access is expected to remain central. The outcomes from initiatives like Patch the Planet will provide valuable data for shaping future directions in this area.

The trajectory suggests that AI will become an integral part of security practices, changing how vulnerabilities are managed across the software supply chain. Organizations will need to develop new skills to work effectively with these tools, combining human expertise with AI outputs for optimal results. The continued focus on open source security through initiatives like Patch the Planet indicates a commitment to protecting the digital commons. This long-term perspective is essential for the sustainable development of AI technologies in the cybersecurity domain.

## Sources

1. [OpenAI announced the expansion including full GPT-5.5-Cyber release, Codex Security plugin update, Patch the Planet, and the 85.6% CyberGym score, along with Codex Security scanning over 30 million commits across more than 30,000 codebases with more than 70,000 findings marked fixed.](https://openai.com/index/daybreak-securing-the-world/)
2. [The partnership with OpenAI resulted in hundreds of discovered bugs, 64 pull requests, 51 issues filed across 19 projects in the first week, with 37 patches merged.](https://blog.trailofbits.com/2026/06/22/introducing-patch-the-planet/)
3. [Industry expert Anthony Grieco commented on the role of frontier models in cybersecurity operations.](https://openai.com/index/gpt-5-5-with-trusted-access-for-cyber/)

---
Source: https://aiintelreport.com/frontier-models/openai-gpt-5-5-cyber-daybreak-expansion
Index: https://aiintelreport.com/llms.txt · Full text: https://aiintelreport.com/llms-full.txt