OpenAI

Cybersecurity · Artificial Intelligence

Launch

OpenAI launches GPT-5.6-Cyber for vetted security researchers

Restricting access reflects how thin the line has become between finding a vulnerability and exploiting it.

• OpenAI launched GPT-5.6-Cyber, a variant of GPT-5.6 Sol tuned for offensive-security work, as part of an expanded Daybreak program for vetted researchers.
• The model handles vulnerability discovery, exploit-chain development, authentication bypass, and privilege escalation for authorized penetration testers.
• Daybreak now splits into two tiers: Daybreak Blue gives general defenders GPT-5.6 Sol, while Daybreak Red restricts GPT-5.6-Cyber to vetted security firms.
• In OpenAI's own testing, GPT-5.6-Cyber answered 95% of advanced cyber requests versus just 1.5% for the standard GPT-5.6 Sol model.
• The launch came three days after OpenAI disclosed it halted parts of its upcoming Astra model after finding it could independently execute cyberattacks, crossing a 'critical' risk threshold.
• The move follows Anthropic's earlier release of Mythos, a rival cyber-focused model, as AI labs race to arm defenders before attackers weaponize the same capabilities.
• Gating the most capable cyber model behind a vetted-researcher tier signals AI labs now treat raw model capability itself, not just its outputs, as a dual-use security risk.