OpenAI has recently launched an initiative known as Daybreak, designed to leverage its advanced large language models (LLMs) and its AI-coding assistant, Codex, to assist developers in creating secure software from the very foundation. This announcement, made on May 12, marks a significant step in the ongoing evolution of AI and cybersecurity, aiming to proactively address vulnerabilities within software during the initial stages of development.
Daybreak is an extension of OpenAI’s Trusted Access for Cyber (TAC) program, a framework that limits access to specific frontier models to a select group of organizations. This initiative is timely as cyber threats continue to grow, prompting the need for more robust defenses within the software development lifecycle. By integrating its new LLMs with a focus on security, OpenAI seeks to equip developers with sophisticated tools that can not only identify vulnerabilities but also rectify them effectively.
The initiative is anchored by three of OpenAI’s leading models: the general-purpose version of GPT‑5.5; GPT‑5.5 with TAC, which provides enhanced security measures for verified defense operations in controlled environments; and GPT‑5.5‑Cyber, designed specifically for cybersecurity. In addition to these models, Daybreak includes Codex Security, a code review assistant powered by Codex. This tool is currently available as a research preview and offers capabilities that are invaluable for developers aiming to enhance the security of their code.
Whereas the TAC program primarily supports vetted users in utilizing LLMs to pinpoint and resolve vulnerabilities, Daybreak addresses the challenge of software security even earlier in the development process. By emphasizing prevention and proactive measures, this initiative could potentially reshape how software is built, ensuring that security is integrated into the design rather than simply addressed post-production.
OpenAI has also shared insights into the diverse tasks that software developers and cybersecurity professionals can accomplish as part of the Daybreak initiative through a series of short videos linked on social media platforms. The capabilities showcased include scanning a codebase with Codex Security’s ten subagents, which can identify vulnerabilities, fix them, and implement regression tests. Additionally, the initiative allows users to manage a backlog of vulnerabilities by prioritizing issues based on severity and exploitability and deploying agents to open relevant pull requests. Perhaps most notably, the automation of vulnerability detection, validation, and response represents a groundbreaking advancement—enabling organizations to stay ahead of the curve by continuously monitoring for the latest Common Vulnerabilities and Exposures (CVEs) and scrutinizing logs for evidence of exploitation.
The overarching goal of the Daybreak initiative, as highlighted in OpenAI’s announcement, is clear: to accelerate the efforts of cybersecurity professionals and to ensure ongoing security in software development. This alignment of capability expansion with principles of trust, verification, proportional safeguards, and accountability underscores OpenAI’s commitment to responsible AI use. Recognizing the intrinsic risks associated with powerful AI tools, OpenAI aims to foster an ecosystem where such technologies can be harnessed for good without compromising security.
Additionally, OpenAI has signaled plans to deploy new “cyber-capable models” in collaboration with industry leaders and government entities, reinforcing the collaborative nature of this venture. As of May 2026, the TAC program boasts participation from hundreds of organizations and thousands of individual cybersecurity defenders. The diverse lineup includes major IT and cybersecurity corporations such as Akamai, Cisco, Cloudflare, and CrowdStrike, as well as important financial institutions like Bank of America and Goldman Sachs.
While currently collaborating with only a few government-linked research organizations, including the US Center for AI Standards and Innovation (CAISI) and the UK AI Security Institute (UK AISI), OpenAI expressed intentions to broaden these collaborations to include more government agencies in the near future.
Anthony Grieco, Senior Vice President and Chief Security and Trust Officer at Cisco, shared his insights on the significance of frontier models like GPT-5.5. He noted that these models act as “powerful force multipliers for defenders,” fundamentally transforming operational speed and efficiency. Grieco emphasized the importance of trust alongside speed, arguing that the true value of this technology lies not merely in the models themselves but in the comprehensive framework that surrounds them. This framework is essential for creating more secure products, positioning innovation and reliability as two sides of the same coin.
As OpenAI forges its path into the realm of cybersecurity with the Daybreak initiative, it heralds a new chapter in the integration of AI technologies within secure software development practices. With ongoing collaborations and a focus on proactive measures, OpenAI aims to redefine how vulnerabilities are addressed, ultimately advancing security in an increasingly digital world.