OpenAI's ChatGPT Cyber Model Escalates AI-Powered Vulnerability Discovery
OpenAI released ChatGPT Cyber for offensive security researchers, but experts warn the speed of AI-assisted exploit discovery now outpaces defenders' ability to respond.
Last verified:
AI Model Finds Critical Ubuntu Exploit in Hours
According to TechCrunch, Max Justicz, chief technology officer and co-founder of robotics startup Charge Robotics, used an OpenAI model to identify a severe vulnerability in Ubuntu’s Linux kernel—one that, if exploited, could grant attackers full control of affected servers. Justicz leveraged GPT 5.6 Sol, accessed through OpenAI’s Daybreak Blue defensive security program, and obtained the exploit in approximately two hours using only a conversational prompt referencing a previously fixed bug in the same codebase. He subsequently reported the vulnerability to the Debian Project, which maintains the software and released a patch; no public evidence of exploitation exists.
OpenAI’s Dual Cybersecurity Programs
According to TechCrunch, OpenAI has separated its cybersecurity research access into two tiers. Daybreak Blue, the defensive track, provides cybersecurity defenders access to frontier models for vulnerability discovery and remediation. Daybreak Red, the newly emphasized offensive track, provides attackers and red-team researchers with ChatGPT Cyber to hunt for exploits and develop attack techniques. The distinction signals OpenAI’s acknowledgment that frontier model capabilities now enable both defense and offense at speeds previously unattainable.
The Defense-Offense Speed Asymmetry
According to TechCrunch, Justicz warned that the ease of AI-assisted exploit discovery far outpaces human defenders’ patching capacity. He wrote in a social media thread that frontier models, when aimed at unaudited code or binaries, represent “huge alpha”—significant competitive advantage—for attackers. TechCrunch reports that Justicz’s observation reflects a broader concern among cybersecurity professionals: while defenders must manually audit, test, and coordinate patches across diverse systems, attackers can now leverage AI models to systematically scan vast codebases for exploitable flaws within hours or days.
Why This Matters
The gap between vulnerability discovery speed and patch deployment has become the dominant asymmetry in cybersecurity. If frontier models continue to accelerate exploit discovery while patch deployment remains bottlenecked by human review, testing, and distribution cycles, defenders face a structural disadvantage that money and tools alone may not close. Organizations maintaining open-source infrastructure—Linux distributions, browsers, kernel maintainers—will need to either accelerate their triage workflows or restrict access to frontier models before exploit discovery scales beyond their capacity to respond. OpenAI’s decision to create separate defensive and offensive tiers suggests the company recognizes this risk but remains committed to enabling offensive research, betting that the net benefit of coordinated disclosure outweighs the risk of acceleration in the exploit marketplace.
Frequently Asked Questions
What is ChatGPT Cyber and who has access to it?
According to TechCrunch, ChatGPT Cyber is OpenAI's new cybersecurity-focused model available to users in the Daybreak Red program, which is designed for offensive cybersecurity researchers conducting vulnerability hunting and exploit development.
How did Max Justicz find the Ubuntu vulnerability?
Justicz used GPT 5.6 Sol, an earlier OpenAI model accessed through the Daybreak Blue defensive program, and simply asked it to find bugs as severe or worse than a previously discovered vulnerability in the same codebase. The model returned a complex exploit in approximately two hours.
Does the speed of AI-assisted discovery outpace defense?
Yes. Security experts quoted by TechCrunch argue that frontier models are making it easier for even novice hackers to find exploits, while defenders struggle to patch vulnerabilities faster than they are discovered.
What happened to the Ubuntu vulnerability after discovery?
Justicz responsibly reported the issue to the Debian Project, which maintains the affected software and issued a patch. There has been no known evidence of exploitation in the wild.