OpenAI2026-08-12 07:45:16OpenAI unveils GPT-5.6-Cyber and says the security model reaches a 95% task completion rateOpenAI said on Aug. 10 that it had launched GPT-5.6-Cyber, a model built for advanced cybersecurity work, and restructured its Daybreak program into two access tiers. In OpenAI’s description, the model can complete about 95% of assigned high-risk dual-use exploit tasks, including attack-chain assembly, authentication bypass, and privilege escalation. The company contrasted that with GPT-5.5-Cyber, which it said posted a 57.3% completion rate on the same class of work. OpenAI also said the newer system identified previously unknown flaws in real software, including two chainable vulnerabilities in Chrome’s V8 engine that could corrupt memory and break out of the V8 sandbox. The company noted that the 95% figure measures whether the model chooses to answer, not whether every answer is correct, a point that has drawn outside discussion. Access is being tightly controlled: Daybreak Blue is intended for broader defensive use, while Daybreak Red covers advanced vulnerability research, validation, and security testing, with entry gated by identity checks, account security requirements, behavior monitoring, approved-use restrictions, and legal commitments.1630
OpenAI2026-08-10 23:34:16OpenAI expands Daybreak and launches GPT-5.6-Cyber for authorized security researchOpenAI said on Aug. 11 that it has expanded its cybersecurity defense program, Daybreak, and introduced GPT-5.6-Cyber, a model built for cybersecurity work. The company said the model is intended for authorized security researchers and defense teams to improve vulnerability discovery, threat analysis, and security testing. OpenAI framed the move as a response to attackers using AI to carry out faster and broader cyberattacks, arguing that defenders need access to advanced AI capabilities earlier. Daybreak now has two access tiers. Daybreak Blue is aimed at most defense teams and offers general-purpose models including GPT-5.6 Sol for vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. Daybreak Red is designed for advanced security research and provides access to GPT-5.6-Cyber for authorized vulnerability research, validation, and security testing. According to OpenAI, GPT-5.6-Cyber is trained on GPT-5.6 Sol and optimized for tasks such as finding zero-day vulnerabilities and analyzing exploit chains. The company said the model has already been used in real-world vulnerability research and identified high-severity flaws in software including Google Chrome’s V8 JavaScript engine. OpenAI also said it helped uncover high-severity issues in mobile operating systems, databases, and operating system kernels.2010
OpenAI2026-06-23 03:07:48OpenAI Launches GPT-5.5-Cyber Full Version, Surpasses Mythos 5 on CyberGym BenchmarkOpenAI has officially released the full version of GPT-5.5-Cyber for cybersecurity defense. The model scored 85.6% in the CyberGym benchmark, outperforming GPT-5.5 and Anthropic's Mythos 5. The company also upgraded Codex Security to automate vulnerability patching and launched the "Patch the Planet" open-source security initiative with partners.510
OpenAI2026-06-23 03:07:48OpenAI Releases Full GPT-5.5-Cyber Model, Outperforms Mythos 5 on CyberGym BenchmarkOpenAI has launched the full GPT-5.5-Cyber model for cybersecurity defense, achieving an 85.6% single-model score on the CyberGym benchmark, surpassing GPT-5.5 (81.8%) and Anthropic Mythos 5 (83.8%). The company also upgraded Codex Security, launched the "Patch the Planet" open-source security initiative, and onboarded partners like Palo Alto Networks and Wiz for the Daybreak program.490
