OpenAI Expands Cybersecurity Program 'Daybreak'

Release of GPT-5.5-Cyber and Codex Security

OpenAI is expanding its Daybreak initiative, which leverages state-of-the-art artificial intelligence (AI) models to strengthen cybersecurity capabilities.

OpenAI Daybreak GPT-5.5-Cyber CyberGym Benchmark Performance Comparison. OpenAI.

On June 23, OpenAI announced the official release of GPT-5.5-Cyber and the Codex Security plugin, marking an expansion of the Daybreak initiative.

The goal of Daybreak is to accelerate the entire process from software vulnerability discovery to verification, risk assessment, patch development, testing, and deployment.

OpenAI has released the official version of GPT-5.5-Cyber to certified experts. This version is designed to help discover and fix software vulnerabilities while maintaining general intelligence and the ability to perform complex tasks over extended periods. OpenAI stated that the model achieved 85.6% on the CyberGym cybersecurity benchmark, surpassing GPT-5.5's 81.8% performance.

Codex Security understands enterprise code and threat models, generates threat models, and identifies potential vulnerabilities. Developers and security professionals then decide whether to further investigate and apply patches to the vulnerabilities found by Codex Security.

OpenAI has also launched the Daybreak Cyber Partner Program, which supports selected partners in integrating the trust-based approach of GPT-5.5 and cybersecurity into products and services for their customers.

The company has introduced the "Patch the Planet" program, which enables open-source project maintainers to discover and fix vulnerabilities. This initiative helps professional security researchers verify and patch vulnerabilities in open-source projects, with more than 30 open-source projects expressing interest in participating.

OpenAI is working with governments and organizations around the world to enhance cyber defense capabilities. Since last month, the Korean government has officially been participating in OpenAI’s Government and Trusted Agency Access Program (GTAC). Over the past month, OpenAI has also established collaboration frameworks with Australia, Canada, France, Germany, Japan, and EU institutions.

This content was produced with the assistance of AI translation services.