OpenAI expanded its Daybreak cybersecurity research program with two access tiers: Daybreak Blue (GPT-5.6 Sol with relaxed safeguards for defensive security work like vulnerability discovery and malware analysis, 2% approval rate) and Daybreak Red (specialized GPT-5.6-Cyber model with 95% approval rate for vetted senior security researchers). The company demonstrated the model's effectiveness by discovering two high-severity vulnerabilities in Chrome's V8 engine and responsibly disclosing them to Google. This tiered approach addresses security analysts across skill levels seeking to bypass standard model restrictions for legitimate defensive research.