OpenAI's cyber model finds bugs it can't patch alone

A hobbyist's curiosity, a frontier model, and two hours. That combination was enough for Max Justicz to turn up a severe flaw in code that underpins a widely used open source operating system — a flaw he says a human would have struggled to find. The report landed as OpenAI began pushing its own cybersecurity offerings toward the researchers who defend that same infrastructure.

Justicz, chief technology officer and co-founder of the startup Charge Robotics, told TechCrunch he found the issue using GPT 5.6 Sol, a model released in July, through OpenAI's Daybreak Blue — a program that gives cybersecurity defenders access to powerful frontier models. He reported it to the Debian Project, which maintains the software in question and issued a patch.

There has been no known evidence of exploitation. The finding is notable less for any damage done than for how little effort it took.

Two hours at the prompt

Justicz came to the exercise with prior knowledge of the codebase: he had previously found a vulnerability in the same body of code. Rather than build tooling or spend weeks on manual review, he aimed the model at material he already understood and asked it to go further.

“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 two hours with this,” Justicz said. “It’s extremely convoluted and hard for a human to find, but it works!”

The vulnerability, as Justicz described it to TechCrunch, could allow hackers to take control of servers running Ubuntu, a widely used open source Linux-based operating system.

He did not treat the result as a one-off. In a social media thread about the finding, he argued that the approach generalizes — and that others will reach the same conclusion soon.

“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,” Justicz wrote in a social media thread discussing his finding. “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.”

Justicz said he personally spent about $5,000 on subscriptions and tokens to find several exploits.

ChatGPT Cyber enters the room

OpenAI appears to have noticed the pattern. On November 5, 2026, the company announced a new model focused on cybersecurity, ChatGPT Cyber, would be made available to the highest tier of users in its cybersecurity program, Daybreak Red — the track that gives offensive cybersecurity researchers tools for vulnerability hunting and development.

The announcement followed OpenAI's own account of what the underlying model could do at scale. The day before, the company said the same model had been used to scan and prepare patches for vulnerabilities in the Linux kernel, finding 948 potential issues. OpenAI also said it found 81 “novel” vulnerabilities in Firefox, the web browser.

The company framed the expansion as a matter of reach rather than spectacle.

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

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

Free patches, paid tokens

The economics of this kind of bug hunting are not trivial. Eric Wallace, a researcher on OpenAI's technical staff, described the company's posture toward open source work.

“We are now allocating a significant amount of our frontier compute towards discovering and patching vulnerabilities in open source codebases, fully pro bono.”

— Eric Wallace, researcher on OpenAI's technical staff

By one measure, though, the commitment is modest relative to the company's other research bets. The Linux bug hunt cost $1.2 million, which is significantly less than the approximate $6 million token spend OpenAI invested in finding a potential solution to a famous problem of fluid dynamics.

That comparison sits at the center of the question TechCrunch put to the announcement: whether the resources devoted to defense match the scale of the problem the models are exposing.

A gate that cuts both ways

The Daybreak program has drawn its own critiques, and Justicz is among those raising them. He noted that the frontier lab appears to have “nerfed” Daybreak Blue, making it less useful at finding vulnerabilities.

There is a defensible reason for restraint. Justicz said he joined the program simply by uploading a photo of his driver's license — a light check on who gets access to models capable of serious vulnerability research. Limiting access may keep potential bad actors away from the most capable tools.

The tradeoff is that defenders operate under the same limits. Cybersecurity professionals argue they have one hand tied if they can't use frontier technology while bad actors can deploy open-weight models with abliterated guardrails — models stripped of the safety behavior their makers built in.

The numbers, as OpenAI describes them, are lopsided. The company says just “thousands” of cybersecurity professionals are participating in the Daybreak program, versus a far larger number of potential hackers.

What the numbers add up to

OpenAI has insisted there is a narrow window of time for the people who safeguard digital infrastructure to act before AI models become even better at hacking and easier to access. The figures from its recent work sketch the shape of that window:

• 948 potential issues found in the Linux kernel, with patches scanned and prepared
• 81 “novel” vulnerabilities reported in Firefox
• $1.2 million — the cost of the Linux bug hunt
• $6 million — the approximate token spend OpenAI put toward a fluid dynamics problem, for comparison
• $5,000 — what Justicz says he spent on subscriptions and tokens to find several exploits
• $1 billion — OpenAI's announced initiative for cybersecurity professionals working on critical infrastructure
• “Thousands” — the number of cybersecurity professionals OpenAI says participate in Daybreak

For cyber defenders, the experience can feel less like a windfall than a mountain of new work arriving faster than anyone can staff for it. Every potential issue the models surface still needs a human to triage it, confirm it, and write or review a patch.

The threat that hasn't materialized

It is worth being precise about what has and hasn't happened. The Ubuntu-related flaw Justicz found was reported to Debian, which maintains the software, and a patch was issued. There has been no known evidence of exploitation.

The Linux kernel findings were described by OpenAI as potential issues, with the model used to scan for them and prepare patches — not as confirmed, exploited vulnerabilities. The Firefox figure refers to “novel” vulnerabilities as characterized by OpenAI. The company's statements about model capability come from the lab that built the model and stands to benefit from its adoption.

What is documented is the discovery process itself: a model pointed at unaudited code, a short turnaround, and a result that the person who asked for it found notable for being convoluted enough to elude human review.

Where defenders are left standing

The pitch behind ChatGPT Cyber and the expanded Daybreak tiers is that the same capabilities that make models useful to attackers can be turned toward finding and fixing flaws first. OpenAI's own framing sets the clock: a narrow window before models get better at hacking and easier to obtain.

The critique from the defensive side is not that the models don't work. It's that access is uneven and the capacity to act on what the models find is thinner than the output. A defender with a $5,000 token budget and a license photo can now generate findings that once required specialist expertise and long hours. What that defender, or an overwhelmed maintainer, does with a queue of hundreds of potential issues is a separate problem — one the models don't solve.

OpenAI says the work on open source is pro bono and points to the $1 billion initiative for critical infrastructure as evidence of commitment. Whether that is enough depends on how fast the discovery side outpaces the patching side, and on how many of those “thousands” of participating defenders can keep pace.