Who can access GPT-5.6-Cyber?

Access is restricted. GPT-5.6-Cyber ships through Daybreak Red, a vetted tier for approved defenders doing advanced security work — not general ChatGPT users. OpenAI gates it because the same capability used to patch systems can be used to attack them.

What has GPT-5.6-Cyber found?

OpenAI researchers used GPT-5.6-Cyber to uncover two previously unknown vulnerabilities in V8, the JavaScript engine that powers Google Chrome, as of August 2026.

Quick Answer

What Is GPT-5.6-Cyber? OpenAI Security Model (Aug 2026)

Q: What is GPT-5.6-Cyber?

GPT-5.6-Cyber is OpenAI's cybersecurity-specific model, announced August 10, 2026 and built on GPT-5.6 Sol. It is trained to find zero-day vulnerabilities and develop exploit chains, and it refuses far less often on high-risk dual-use security work. It completes 95% of exploit-development requests on OpenAI's internal eval versus 1.5% for the consumer model.

Published: August 12, 2026

The Short Answer

GPT-5.6-Cyber is OpenAI’s cybersecurity-specific model, launched August 10, 2026 through the Daybreak Red tier for vetted defenders. Built on GPT-5.6 Sol, it finds zero-day vulnerabilities and builds exploit chains, and it completes 95% of exploit-development requests on OpenAI’s internal advanced-cyber eval — versus 1.5% for the consumer model.

What Changed

Standard ChatGPT refuses almost all offensive-security requests. GPT-5.6-Cyber flips that on purpose for a narrow, vetted audience: it’s tuned to answer high-risk dual-use questions so defenders can find and patch flaws before attackers do. OpenAI describes the release as “expanding Daybreak as the cyber-defense window narrows.”

Key Facts (as of August 2026)

	GPT-5.6-Cyber
Vendor	OpenAI
Base model	GPT-5.6 Sol
Announced	August 10, 2026
Access tier	Daybreak Red (vetted defenders)
Internal eval	95% exploit-dev completion (consumer model: 1.5%)
Proven finds	2 unknown vulns in Chrome’s V8 engine

Why It Matters

GPT-5.6-Cyber is the clearest sign yet that frontier labs treat offensive-security capability as a gated product, not a refusal wall. The release followed OpenAI saying it could not rule out a Critical cyber classification in its forthcoming Astra model — which paused some internal testing. The bet: put the capability in defenders’ hands under vetting before it leaks to attackers anyway.

The Reality Check

The 95% figure is OpenAI’s internal number, and access is deliberately hard to get. For everyone else, the consumer models still refuse. The real story is the policy shift — a lab shipping a model designed to answer the questions its flagship refuses, for a whitelist of defenders.

Sources

OpenAI — Expanding Daybreak as the cyber-defense window narrows: openai.com
Implicator — OpenAI’s cyber model at 95%: implicator.ai
eesel AI — GPT-5.6-Cyber: what it is and who can get it: eesel.ai