AI
OpenAI's New Cybersecurity Model Sparks Questions About Whether Security Efforts Keep Pace With AI Speed
As OpenAI expands its cybersecurity capabilities and proves AI can find complex vulnerabilities faster than humans, defenders worry the company isn't doing enough to help them catch up.
TechCrunch — AI
A startup CTO recently demonstrated just how effective AI has become at uncovering security flaws. Using OpenAI's GPT 5.6 Sol model through the company's Daybreak Blue program, he identified a critical vulnerability in Ubuntu's codebase in just two hours — a bug so intricate that human security auditors would likely miss it. The discovery, which could have granted attackers control over Linux servers running the operating system, underscores an uncomfortable truth: artificial intelligence is now better at finding exploitable weaknesses than the defenders meant to stop them.
OpenAI's Expanding Security Arsenal#
OpenAI has responded to this reality by launching ChatGPT Cyber, a specialized model for offensive security research available through its Daybreak Red initiative. The company has also demonstrated the model's capacity by scanning the Linux kernel and identifying 948 potential vulnerabilities, along with 81 previously unknown flaws in Firefox. In tandem, OpenAI announced a $1 billion fund to support cybersecurity professionals using the platform to protect critical infrastructure like power grids and water systems.
"Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers," OpenAI stated.
The Growing Imbalance#
Despite these initiatives, security experts and researchers express skepticism about whether the efforts are sufficient. The volume of potential vulnerabilities AI can uncover appears to vastly exceed the defensive capacity of the cybersecurity community, many of whom work with limited budgets and staffing. One researcher noted he spent approximately $5,000 on subscriptions and computational resources to uncover multiple exploits — a cost many individual defenders or smaller organizations cannot afford.
A particular concern centers on access imbalances. While OpenAI has restricted Daybreak Blue's capabilities (reportedly to reduce misuse risk), the company notes that only thousands of cybersecurity professionals participate in these controlled programs. Meanwhile, malicious actors can deploy unrestricted open-weight AI models with removed safety guardrails, potentially giving them an advantage in the asymmetric race.
Resource Allocation Questions#
OpenAI claims it is dedicating significant computational resources to this work at no cost. However, some observers question the scale of commitment. The Linux vulnerability scanning initiative required $1.2 million in token spending — notable, but dwarfed by the approximately $6 million the company spent investigating fluid dynamics research. Given the stakes of digital infrastructure security, this comparison raises whether the frontier labs are prioritizing the problem adequately.

