OpenAI Ships GPT-5.6-Cyber Through Gated Daybreak Red Access Tier

OpenAI has released GPT-5.6-Cyber, a cybersecurity-specific model that gives approved defenders tools for exploit development and vulnerability research.

The company positioned the launch around a shrinking window for defense. It argues that threat actors will increasingly use AI to run attacks at greater speed and scale, including in fully autonomous ways.

OpenAI Gives Trusted Defenders Frontier Cyber Tools

GPT-5.6-Cyber is built on GPT-5.6 Sol, OpenAI's standard model. Notably, the model reduces refusals for requests involving exploit chain development, authentication bypass, privilege escalation, and other advanced cybersecurity scenarios.

OpenAI says it completes 95% of these requests, compared with 1.5% for its general-purpose GPT-5.6 Sol model.

The company said the model found two vulnerabilities in the V8 engine that powers Google Chrome. It reported them to Google as CVE-2026-15903. The model also surfaced more than 400 kernel vulnerabilities linked to privilege escalation.

Meanwhile, OpenAI said it's expanding Daybreak to two access tiers. Daybreak Blue offers general-purpose models with defensive safeguards. Daybreak Red provides access to purpose-trained cybersecurity models, including GPT-5.6-Cyber.

… Continue reading the full article at the original source below.