OpenAI is building an AI Cyber Defence System called Daybreak
OpenAI has launched Daybreak, a new AI cyber defense platform designed to detect vulnerabilities and accelerate software security workflows.
OpenAI is moving deeper into cybersecurity with a new initiative called Daybreak, an AI-powered cyber defence platform designed to help organisations detect vulnerabilities, validate fixes, and secure software before attackers can exploit weaknesses.
The company unveiled Daybreak on 11 May 2026, positioning it as part of a broader push to embed AI directly into software security and cyber defence workflows. According to OpenAI, the platform combines GPT-5.5 models, Codex Security, and specialised cybersecurity safeguards to help defenders identify risks earlier and accelerate remediation.
Building cyber defence into software
OpenAI says Daybreak is based on the idea that cybersecurity should begin during software development rather than after vulnerabilities are discovered in production systems.
According to the AI firm, the platform is designed to help security teams reason across large codebases, analyse unfamiliar systems, identify subtle vulnerabilities, validate fixes, and generate remediation guidance faster than conventional workflows.
OpenAI described the goal as making software “resilient by design” rather than relying only on reactive patching. The initiative arrives at a time when AI-powered cyberattacks are becoming more sophisticated, and security teams are increasingly struggling to keep pace with growing software complexity.
What Daybreak actually does
Daybreak combines several defensive cybersecurity capabilities into one operational system. The platform can perform secure code reviews, generate editable threat models, analyse dependency risks, validate patches, and assist with vulnerability triage.
According to OpenAI, security teams can also use the system to generate and test patches directly inside repositories while maintaining scoped access controls and audit trails. The company added that Daybreak is designed to reduce false positives by validating likely vulnerabilities in isolated environments before escalating issues to human security teams.
ChatGPT maker said it also integrates with detection and remediation workflows so organisations can continuously monitor and track fixes across development systems.
Three levels of cyber access
As part of Daybreak, OpenAI introduced different levels of cybersecurity model access depending on the type of organisation and workflow involved. The default GPT-5.5 configuration includes standard safeguards for general-purpose work.
A second tier, called Trusted Access for Cyber, is intended for verified defensive workflows such as vulnerability analysis, malware review, and secure code validation.
The most permissive option, GPT-5.5-Cyber, is reserved for specialised authorised workflows including penetration testing and controlled red-team exercises. The stricter access controls are designed to balance defensive capability with safeguards against misuse.
Why OpenAI is launching this now
The launch comes amid growing concern around AI-assisted hacking and offensive cyber capabilities. Recent industry reports have shown that AI systems are increasingly capable of identifying vulnerabilities, generating exploit code, and automating portions of cyberattacks.
OpenAI revealed Daybreak is intended to help shift the balance toward defenders by accelerating detection and remediation faster than attackers can exploit flaws. The announcement also places OpenAI in more direct competition with Anthropic, which recently introduced its own cybersecurity-focused initiative called Mythos under Project Glasswing.
Unlike Anthropic’s more restricted rollout, OpenAI is positioning Daybreak as a broader enterprise security platform integrated with Codex Security and multiple GPT-5.5 cyber models.
Security partners are already involved
The AI firm claims several major cybersecurity companies are already working with Daybreak capabilities. The company listed partners, including Cloudflare, CrowdStrike, Cisco, Oracle, Palo Alto Networks, Zscaler, Akamai, and Fortinet, among organisations participating in the initiative.
Dane Knecht, Chief Technology Officer at Cloudflare, said OpenAI’s cyber capabilities could help security teams bring stronger reasoning and more agentic execution into security workflows. OpenAI is also coordinating with industry and government partners as it prepares to deploy more advanced cyber-capable AI systems in the coming months.