- OpenAI launched AI security tools and warned businesses to use them ASAP or risk getting hacked, amid growing concerns over AI-powered cyberattacks.
- OpenAI unveiled an unreleased next-generation cybersecurity-focused model during a livestream, scoring 77.0% on ExploitGym, a benchmark measuring the ability to generate code that exploits vulnerabilities.
- The next-generation model's score of 77.0% far exceeds the 52.6% posted by the current GPT-6 Cyber and roughly 38% for Anthropic's Mythos 5.2.
- GPT-6 Cyber has already delivered results in the practical range, discovering 81 previously unknown vulnerabilities in the Firefox browser at a severity rating of "high", and examining roughly 1,000 vulnerabilities across Linux's vast codebase in a single run.
- OpenAI president Greg Brockman said at a keynote that companies will be secure only to the extent of the frontier AI models they use to defend themselves and how well they utilize them.
- Brockman acknowledged that traditional security practices remain important but said organisations need to rethink their approach in light of what advanced AI models can do.
- The livestream titled "Frontier at the Cyber" was held in the early hours of November 13, 2026, during which the company abruptly revealed part of the performance of its next-generation cybersecurity-focused model.
OpenAI has launched two innovative AI security tools, Daybreak Red Team and Daybreak Defense Factory, aimed at helping businesses identify and fix vulnerabilities before they can be exploited by hackers.
The tools leverage AI agents to test applications for weaknesses, allowing organizations to proactively simulate attacks and address security gaps.
During a recent livestream titled "Frontier at the Cyber", OpenAI unveiled a next-generation cybersecurity model that achieved a remarkable 77.0% score on ExploitGym, significantly outperforming the current GPT-6 Cyber model, which scored 52.6%. This new model's capabilities are crucial as concerns about AI-driven cyberattacks escalate.345678913
OpenAI president Greg Brockman emphasized the need for companies to enhance their security measures, stating, "We're now at a point where companies will be secure only to the extent of the frontier model that they use to defend themselves and how well they utilize that."1011
The Daybreak Red Team tool tests web applications for vulnerabilities, while Daybreak Defense Factory focuses on patching identified weaknesses. Both tools are part of OpenAI's strategy to help businesses stay ahead of increasingly sophisticated cyber threats.
OpenAI's recent performance highlights include discovering 81 previously unknown vulnerabilities in the Firefox browser, showcasing the practical applications of its AI models in real-world scenarios.
As the landscape of cybersecurity evolves, OpenAI's tools represent a significant advancement in the fight against cybercrime.
“The unreleased model's 77% score on ExploitGym nearly doubles Anthropic's Mythos 5.2's ~38%, while GPT-6 Cyber has already found 81 high-severity Firefox flaws. OpenAI president Greg Brockman said companies will be secure only to the extent of the frontier AI models they use to defend themselves.”
