Max Justicz says GPT 5.6 Sol found a serious flaw in code related to Ubuntu systems in two hours, after he asked it to search for issues of similar or greater severity.

If exploited, the flaw could let attackers take control of affected systems; there is no known evidence it was used in an attack.

The reported case offers a concrete example of AI-assisted vulnerability discovery, with limited details about the analysis itself.

How was the problem detected?

Max Justicz, co-founder and CTO of startup Charge Robotics, told TechCrunch that he used GPT 5.6 Sol through OpenAI’s Daybreak Blue. The program gives experts working on system protection access to powerful AI models. GPT 5.6 Sol was released in July.

Justicz had previously identified another problem in the same code, which had been fixed several years earlier. This time, he asked the model to look for problems of equal or greater severity. According to him, the model found the new problem within two hours.

Justicz described the problem as complex and difficult for a human to detect. He also said he spent about $5.000 on subscriptions and using models to identify several problems.

What OpenAI is doing for cybersecurity

OpenAI has announced ChatGPT Cyber, a model for cybersecurity work. It will be available to the highest-level users of Daybreak Red, a program that provides researchers with tools to identify problems and prepare solutions to them.

↳

READ ALSO

HACKS.GR / RELATED

AI Unlocked

Nova Lake may limit speed to 5,1 GHz at 90°C, no confirmation from Intel

↗

A day before the announcement, OpenAI said the same model had examined Linux code and helped prepare fixes for 948 potential issues. The company also said it had identified 81 new issues in Firefox.

OpenAI says it wants to expand Daybreak to include organizations of all sizes, independent researchers, and those who maintain open-source software whose code is available for review and modification. It has also announced a $1 billion initiative for cybersecurity professionals participating in the program who work to protect critical infrastructure, such as power grids and water systems.

There are concerns, however, about whether experts protecting systems have enough access to powerful models, while attackers may be using models that have had security restrictions removed. Justicz said OpenAI appears to have recently limited Daybreak Blue's ability to detect problems. He also said he had to upload a photo of his driver's license to join the program.

OpenAI says that “thousands” of cybersecurity professionals are participating in Daybreak. Eric Wallace, a researcher on the company’s technical staff, said that OpenAI is making a significant portion of its computing power available for free to help identify and fix issues in open source projects.

The effort to control Linux cost $1,2 million. OpenAI had spent about $6 million using models to find a possible solution to a well-known scientific problem involving the movement of liquids and gases.

>Stay connected with Hacks.gr

Set us as a preferred source on Google and find us on social media.

Preferred source

↗

end of article

Read the original English article on Hacks.gr