Breaking

OpenAI Expands Daybreak With Codex Security and GPT-5.5-Cyber Release

June 22, 2026 at 13:05 EDT

Security
AI Agents
Software Dev & Coding

On June 22, 2026, OpenAI announced an expansion of OpenAI Daybreak, its initiative for cyber defenders, unveiling Codex Security, a vulnerability plugin built into Codex; the full version of GPT-5.5-Cyber, a cyber-specific model; a Cyber Partner Program for security firms; and Patch the Planet, an effort to fix open-source vulnerabilities. The company says the goal is to patch vulnerable software at "machine speed" and broaden defenders' access to AI.

June 22, 2026 · OpenAI

Daybreak Expands: Patching Vulnerabilities at "Machine Speed"

OpenAI extends its cyber-defense initiative with a Codex Security plugin, the full GPT-5.5-Cyber model, and a Cyber Partner Program — aiming to tilt the balance from attackers toward defenders.

XBOW benchmark — miss rate (lower is better)

GPT-5.5 misses fewer vulnerabilities than its predecessor — a sharp drop from 40% down to 10%.

40%

GPT-5

10%

GPT-5.5

4×

Fewer misses vs GPT-5 (40% → 10%)

18%

Improvement over Claude Opus 4.6

TAC

Trusted Access for Cyber — permissioned access only

The agentic AppSec workflow

Scan

Full-codebase detection

→

Validate

Threat model + sandbox

→

Patch

Generate fix + export

→

Human Review

Evidence verification

Promise

"Very good" code-level vulnerability discovery
Sandbox validation cuts false positives
Folds AppSec into the development loop

Open Questions

Permissioned access limits & high cost
No governance over autonomous patching
Rival Mythos leads on some benchmarks

Patch the Planet

A push to fix open-source vulnerabilities from discovery to merged fix — with human review at the core.

Partners including Trail of Bits, HackerOne and Check Point embed GPT-5.5 with Trusted Access for Cyber. Pricing and supported environments remain undisclosed.

Continue reading

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