OpenAI Launches GPT-5.5-Cyber and Codex Security β€” Daybreak vs Anthropic Glasswing

OpenAI just launched the full version of GPT-5.5-Cyber β€” its cybersecurity-specific model β€” expanded its Daybreak program with a Codex Security plugin, and launched “Patch the Planet” to fix open-source vulnerabilities at scale. This is OpenAI’s direct answer to Anthropic’s Project Glasswing. The cyber defense arms race between the two largest AI labs is now official.

OpenAI Daybreak β€” The Numbers

85.6%

GPT-5.5-Cyber on CyberGym benchmark

81.8%

Standard GPT-5.5 on same benchmark

8

Security vendor partners in Trusted Access

3

New launches today: Cyber model + plugin + Patch the Planet

What Launched

OpenAI expanded its Daybreak cybersecurity program with three simultaneous releases:

Model

GPT-5.5-Cyber β€” Full Version

A cybersecurity-permissive variant of GPT-5.5, specifically trained to be more permissive on defensive security tasks. Scores 85.6% on CyberGym vs 81.8% for standard GPT-5.5. Limited release to trusted defenders only β€” not available to the general public.

Tool

Codex Security Plugin

Find, validate, and fix vulnerabilities directly inside Codex β€” OpenAI’s AI coding agent. Turns Codex from a code generator into an application security platform. Builds threat models, hunts vulnerabilities, suggests patches.

Initiative

Patch the Planet β€” Open Source Security

A program to find and fix vulnerabilities in widely used open-source software at scale. AI-powered scanning + automated patch generation for the software infrastructure the internet runs on.

The Structural Read β€” Daybreak vs Glasswing

This is OpenAI’s direct answer to Anthropic’s Project Glasswing. The parallels are exact:

Cyber Defense Programs β€” Head to Head

Anthropic β€” Glasswing

Model: Mythos Preview (raw)

Partners: 200 organizations

Results: 23,019 vulns found

Status: Still running despite Fable 5 shutdown

OpenAI β€” Daybreak

Model: GPT-5.5-Cyber (85.6%)

Partners: 8 security vendors

Tool: Codex Security plugin

New: Patch the Planet (open source)

The key insight: Both labs have concluded the same thing β€” cybersecurity is the use case where frontier AI capability is most immediately valuable AND most politically defensible. Glasswing survived the US government shutdown of Fable 5. Daybreak positions OpenAI for the same protected status. The Permission Layer is becoming a cyber defense credential.

CODEX BECOMES A SECURITY PLATFORM

The Codex Security plugin is the strategic move. It embeds vulnerability scanning into the developer workflow β€” the same IDE where code is written. If developers use Codex to write code AND scan it, OpenAI owns both sides of the platform. Write the bug and find the bug in the same tool.

PATCH THE PLANET IS THE PUBLIC GOODS PLAY

Fixing open-source vulnerabilities at scale is good security AND good politics. Governments want AI companies doing socially valuable work. OpenAI fixing the software the internet runs on is a Permission Layer investment β€” the same logic that made Glasswing survive the Fable 5 shutdown.

THE CYBER RACE IS THE MODEL RACE’S SHADOW

Anthropic: Glasswing (200 orgs, Mythos). OpenAI: Daybreak (8 vendors, GPT-5.5-Cyber). Both are competing for government trust through cyber defense. The lab that becomes the default security partner for critical infrastructure gets the one thing money can’t buy: regulatory protection.

The Bottom Line

OpenAI and Anthropic are now running parallel cybersecurity programs β€” Daybreak vs Glasswing β€” each trying to become the AI lab that governments trust with defensive security. GPT-5.5-Cyber at 85.6% on CyberGym, the Codex Security plugin embedding vulnerability scanning into developer workflows, and Patch the Planet fixing open-source bugs at scale. This is not a product launch. It is a Permission Layer strategy β€” the same playbook that kept Glasswing running when the US government shut everything else down.

Business Engineer Framework

The AI Supercycle β€” The Cyber Defense Layer

Read the AI Supercycle β†’

Sources: OpenAI Daybreak, OpenAI on X, MLQ β€” June 22, 2026

Scroll to Top

Discover more from FourWeekMBA

Subscribe now to keep reading and get access to the full archive.

Continue reading

FourWeekMBA