Breaking

OpenAI Ships Full GPT-5.5-Cyber and Tools to Speed Up Patching

June 22, 2026 at 15:31 EDT

Foundation Models
Security
AI Agents

OpenAI on June 22, 2026 released the full version of its cybersecurity-focused large language model, GPT-5.5-Cyber, along with new tools designed to accelerate the patching of vulnerabilities. Announced via the company blog post "Daybreak: Tools for securing every organization in the world," the model is offered through a limited release under the Trusted Access for Cyber (TAC) program for verified defenders. The release also included an update to Codex Security, a vulnerability scanning and patch-generation plugin integrated into the Codex coding AI, and the Patch the Planet initiative to support open-source vulnerability fixes.

April 2026 · OpenAI

GPT-5.5-Cyber Goes Full Release — Built to Patch, Not Just Find

A cyber-focused, identity-verified variant for vetted defenders of critical infrastructure — now state-of-the-art on the CyberGym benchmark and shipped with new tools, Codex Security and Patch The Planet, aimed at automated remediation.

71.4%

Avg. pass rate on expert-level AISI cyber tasks

State-of-the-art on CyberGym, ahead of Mythos 5

>50%

Of cases Sophos resolved without human intervention

Access tiers — increasing flexibility, increasing verification

Default

GPT-5.5
Standard safeguards

TAC

GPT-5.5 + TAC
Code review, triage, malware analysis, patch validation

Cyber

GPT-5.5-Cyber
Red teaming, pen-testing, exploit validation + extra monitoring

Codex Security — remediation, not just discovery

Build threat model

→

Explore attack paths

→

Validate in isolation

→

Propose patch

Patch The Planet applies this loop to automatically fix open-source vulnerabilities, powered by the Daybreak platform.

Defenders — positive

Sophos' agentic SOC closed over half of cases autonomously. Cisco, Intel, SentinelOne and Snyk cite it as a force multiplier for speed.

Open question

Routine devops work increasingly trips "cyber risk" flags. The balance between access controls and flexibility for a dual-use tool stays unresolved.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.