OpenAI Unveils New Security-Focused Model 'GPT-5.6 Cyber'...Reorganizes Security Program

Discovered over 400 actual security vulnerabilities

Daybreak program reorganized into Red and Blue tiers

OpenAI has introduced a new model specialized in security and is reorganizing its security program, Daybreak.

On August 11, OpenAI announced the launch of 'GPT-5.6 Cyber,' a cybersecurity-focused model. This model is characterized by its specialization in advanced vulnerability research, exploit verification (the ability to verify the likelihood of vulnerability exploitation), and security testing.

Based on OpenAI's latest model, GPT-5.6 Sol, this new release has been specially trained to reduce unnecessary refusals during security work. As a result, it has been optimized for detecting security vulnerabilities and creating remediation patches.

In 'ExploitGym,' which measures the ability to turn vulnerabilities into code that can be used in real attacks, GPT-5.6 Cyber achieved the highest success rate among all models released by OpenAI to date. It also completed 95% of all requests in the evaluation of advanced cybersecurity task completion.

The model succeeded in finding vulnerabilities in actual software as well. While investigating the V8 JavaScript engine used by the Chrome web browser, OpenAI researchers identified two previously unknown vulnerabilities using this model. After verifying their findings, the OpenAI research team notified Google, and Google subsequently patched the relevant vulnerabilities.

In addition, OpenAI reported that using this model, it discovered five vulnerabilities in mobile operating systems, three in databases, and approximately 400 in widely used OS kernel areas.

OpenAI explained that according to its internal safety standards assessment, GPT-5.6 Cyber was rated as "High," which does not reach the "Critical" level of risk.

Furthermore, OpenAI announced that the Daybreak program, which provides cybersecurity models to individuals and organizations carrying out authorized defensive work, will now be divided into 'Daybreak Blue' and 'Daybreak Red.'

Daybreak Red is a tier for organizations performing professional security work and grants direct access to the security-specialized GPT-5.6 Cyber model. Daybreak Blue is a tier that allows for tasks such as vulnerability detection and malware analysis based on the top-tier general-purpose model, GPT-5.6 Sol.

Both programs will be available only to individuals and organizations carrying out pre-approved security work. To use these services, processes such as identity verification, account security, and monitoring must be completed.

Meanwhile, OpenAI previously decided to slow down the development of its next-generation AI model, Astra, as it determined that the model may have reached a "Critical" risk level according to its own safety standards.

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