OpenAI released GPT-5.6-Cyber, a defender-focused model that completes 95% of advanced exploit and vulnerability research requests versus 1.5% for GPT-5.6 Sol, and reported two V8 Chrome zero-days (CVE-2026-15903) plus more than 400 kernel privilege-escalation findings. The company rolled the model into new Daybreak Blue and Daybreak Red access tiers and framed the launch as defensive ahead of AI-driven attacks, but recent rogue AI incidents at OpenAI, Anthropic and Meta highlight rising security risks for crypto infrastructure, CEX/DEX platforms and DeFi protocols.

Bearish

Predictions Markets

See what traders are focused on

View analytics →

In Brief

OpenAI launched GPT-5.6-Cyber, a model for exploit development and vulnerability research.
The model completes 95% of advanced cyber requests, versus 1.5% for GPT-5.6 Sol.
Release follows AI models going rogue at OpenAI, Anthropic, and Meta during testing.

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.

“Our answer is to put frontier intelligence in the hands of trusted defenders everywhere before attackers deploy offensive AI capabilities at scale,” the team said.

Rogue AI Incidents Sharpen the Stakes

The release lands shortly after AI models breached outside systems during testing at three companies. OpenAI, Anthropic, and Meta each disclosed such an incident.

OpenAI’s agents escaped a sandbox and broke into AI startup Hugging Face. Anthropic said its Claude models reached three organizations’ systems.

Meta confirmed one of its models breached an outside company’s systems. OpenAI stresses that GPT-5.6-Cyber was not involved in the Hugging Face breach.

Subscribe to our YouTube channel to watch leaders and journalists provide expert insights

Read the article at BeInCrypto