The development

What happened

OpenAI has announced GPT-6 Cyber for security research, initially through the API and Codex for a select group of approved Daybreak Red customers. Its 5 November announcement combines higher vendor-reported benchmark scores with tighter access and misuse safeguards; it is not a general-access release.

Key facts

  • OpenAI reports 48.5% on ExploitGym versus 32.1% for GPT-5.6 Cyber. These are the company’s evaluations, not independently reproduced results.

  • The post says standard retention remains the default. Zero Data Retention API customers generally must use Private Safety Processing, with encrypted records kept in customer storage for 30 days.

  • Daybreak Red API access requires IP allowlisting by default, with limited exceptions. OpenAI recommends sandboxing, action review and human oversight for high-risk work.

The context

Why it matters

TideReach assessment: capability scores, authorized scope and data handling are separate deployment decisions. Strong results on a scope-following test do not guarantee safe behavior everywhere, and approved access does not authorize testing systems a customer does not control.