OpenAI Launches GPT-5.6-Cyber: A New AI Model Built for Cybersecurity
Announced on August 10, 2026, a new model arrives as OpenAI warns that the cybersecurity landscape is changing rapidly, with increasingly capable AI potentially allowing attackers to conduct cyber operations at unprecedented speed and scale.
OpenAI has introduced GPT-5.6-Cyber, a cybersecurity-specific AI model designed to help authorized security researchers discover vulnerabilities, validate exploits, and conduct advanced security testing. Announced on August 10, 2026, the model arrives as OpenAI warns that the cybersecurity landscape is changing rapidly, with increasingly capable AI potentially allowing attackers to conduct cyber operations at unprecedented speed and scale.
GPT-5.6-Cyber is built on GPT-5.6 Sol but receives specialized training for complex cybersecurity tasks, including zero-day vulnerability discovery and exploit-chain development. It is being offered through Daybreak Red, OpenAI’s access tier for approved organizations and individuals conducting advanced vulnerability research, exploit development, and red-team activities.
A Major Reduction in Cybersecurity Refusals
One of the most notable changes is GPT-5.6-Cyber’s ability to respond to advanced, dual-use cybersecurity requests that previous models frequently refused. OpenAI’s internal Advanced Cybersecurity Completion Rate evaluation found that GPT-5.6-Cyber completed 95% of tested advanced requests, compared with just 1.5% for GPT-5.6 Sol and 57.3% for GPT-5.5-Cyber.
The company says this change is intended to reduce unnecessary barriers for legitimate security researchers while maintaining strict controls around access and usage. Daybreak access includes identity verification, monitoring, approved-use restrictions, and legal attestations.
Stronger Vulnerability Research Capabilities
OpenAI reports that GPT-5.6-Cyber outperformed GPT-5.6 Sol and GPT-5.5-Cyber on ExploitGym, a controlled evaluation measuring an AI agent’s ability to turn known vulnerabilities into working exploits. The model also performed better on an internal benchmark focused on discovering novel vulnerabilities and accurately assessing their severity.
However, GPT-5.6-Cyber does not outperform GPT-5.6 Sol in every test. On OpenAI’s Vulnerability Discovery and Report Writing evaluation, GPT-5.6-Cyber performed worse, partly because it sometimes generated shorter and less detailed vulnerability reports. On the more demanding ExploitBench evaluation, GPT-5.6 Sol also performed better in the standard 300-turn setting.
Real-World Vulnerabilities Discovered
Perhaps the most significant evidence comes from real-world research. OpenAI says GPT-5.6-Cyber helped identify two previously unknown vulnerabilities in V8, Chrome’s JavaScript engine. Researchers validated the findings and reported them to Google, which fixed one vulnerability as CVE-2026-15903.
OpenAI also reports identifying vulnerabilities across mobile operating systems, databases, and an operating-system kernel, including more than 400 vulnerabilities capable of leading to privilege escalation in the kernel studied.
What This Means for Cybersecurity
GPT-5.6-Cyber represents an important shift toward specialized AI tools for defensive security. OpenAI emphasizes that the model remains below its Critical cybersecurity capability threshold under its Preparedness Framework, despite reaching the High threshold.
The launch ultimately highlights a growing cybersecurity race: as AI becomes more capable of finding and exploiting weaknesses, defenders need equally powerful tools to discover and fix those weaknesses first. GPT-5.6-Cyber is OpenAI’s latest attempt to give trusted defenders that advantage—but its capabilities also underline why careful access controls, sandboxing, monitoring, and human oversight will remain essential.
