OpenAI's New Cybersecurity Model: Arms Race Accelerates

A startup cofounder with a hobbyist's interest in cybersecurity found a severe Ubuntu vulnerability in roughly two hours by pointing an OpenAI model at code he had previously audited. Max Justicz, chief technology officer and cofounder of Charge Robotics, told TechCrunch the bug — if exploited — could have let hackers take control of servers running Ubuntu, a widely used open-source Linux-based operating system.

Justicz used GPT 5.6 Sol, released in July, through OpenAI's Daybreak Blue program, which gives cybersecurity defenders access to powerful frontier models. He had found a vulnerability in the same codebase before.

"I literally just aimed ChatGPT at the same body of code and said 'I found this bug previously that was fixed several years ago, please find bugs as bad or worse,' and it came back in 2 hours with this," Justicz said. "It's extremely convoluted and hard for a human to find, but it works!"

He reported the issue to the Debian Project, which maintains the software and issued a patch. There is no known evidence of exploitation. The ease of the find suggests defenders have seen only the tip of the iceberg of potential exploits that even novice hackers could weaponize with AI models.

What did the researcher warn?

Justicz framed the finding as a preview of what is coming. "I can only imagine what one could do with more time and access to less neutered models. We should expect some extremely crazy bugs to be found and exploited over the next few months," he wrote in a social media thread. "This might be obvious to most folks here, but there is currently huge alpha in taking some code or binary someone has not audited yet and just aiming a powerful AI model at it."

The cost asymmetry is the story. Justicz said he personally spent about $5,000 on subscriptions and tokens to find several exploits — pocket change against the potential damage of a server takeover bug. That figure is the strongest signal yet that serious vulnerability research is now within reach of individuals, not just well-funded teams.

How did OpenAI respond?

OpenAI appears to have noticed. The company announced ChatGPT Cyber, a new model focused on cybersecurity, available to the highest tier of users in its Daybreak Red program, which gives offensive security researchers tools for vulnerability hunting and development.

A day earlier, OpenAI said the same model had scanned and prepared patches for the Linux kernel, finding 948 potential issues. The company also found 81 "novel" vulnerabilities in Firefox, the web browser.

"Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers," OpenAI said in a statement.

Beyond model access, OpenAI announced a $1 billion initiative to support cybersecurity professionals in the program who work on critical infrastructure such as the electrical grid or water systems.

Does the money match the rhetoric?

OpenAI has insisted there is a narrow window for the people safeguarding digital infrastructure to act before AI models become better at hacking and easier to access. The spending record invites scrutiny of that claim.

Eric Wallace, a researcher on OpenAI's technical staff, said: "we are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono."

But the Linux bug hunt cost $1.2 million. That is significantly less than the roughly $6 million in token spend the company invested in finding a potential solution to a famous fluid dynamics problem. In plain terms: OpenAI spent five times more on one mathematics problem than on patching the operating system kernel that underpins much of the internet. Whether that ratio reflects a rational allocation or a mispriced risk is a question the company has not answered.

Is Daybreak keeping defenders ahead?

For cyber defenders, the frontier labs can feel like they have unleashed a mountain of new work that outstrips defenders' resources. The Daybreak program itself draws critiques from its own users.

Justicz notes that OpenAI appears to have recently "nerfed" Daybreak Blue, making it less useful at finding vulnerabilities. On one hand, limiting access may keep potential bad actors out — Justicz said he joined the program simply by uploading a photo of his driver's license, a bar low enough to worry anyone. On the other hand, cybersecurity professionals say they have one hand tied if they cannot use frontier technology while bad actors deploy open-weight models with abliterated guardrails.

The scale gap compounds the problem. OpenAI says just "thousands" of cybersecurity professionals participate in Daybreak, versus a far larger number of potential hackers. A defender-to-attacker ratio measured in thousands against millions is not a comfortable position.

It remains unclear whether frontier labs are devoting enough of their resources to the issue. OpenAI's framing — a narrow window before models improve further — is itself an admission that the model releases driving the risk are outpacing the patching effort meant to contain it.

What is the bottom line?

If the window to patch the digital infrastructure the world depends on is genuinely narrow, more urgency than a $1.2 million bug hunt and a gated model tier may be required.