Breaking

OpenAI Unveils Defense-Focused GPT-5.5-Cyber and Open-Source Patching Drive

June 22, 2026 at 14:37 EDT

Security
Foundation Models
Open Source

On June 22, 2026, OpenAI unveiled GPT-5.5-Cyber, an improved cybersecurity-focused model, and "Patch the Planet," an initiative to use AI to find and fix vulnerabilities in open-source software. Both are offered only to trusted defenders rather than the general public.

June 22, 2026 · OpenAI Daybreak

AI Goes on Defense: GPT-5.5-Cyber & "Patch the Planet"

OpenAI expands its cybersecurity push with a more capable defender-only model and an initiative to find, verify and fix vulnerabilities across open-source software — closing the gap from discovery to patch at machine speed.

projects scanned in a 5-day sprint

pull requests submitted (37 merged)

23-yr

old use-after-free confirmed in OpenBSD

30M+

commits scanned by Codex Security

Benchmarks: GPT-5.5-Cyber vs GPT-5.5

Single-model evaluation scores (% solved) — the new model leads on every task.

85.6%

81.8%

CyberGym

reproducing known bugs

39.5%

26.0%

ExploitGym

bugs into exploits

69.8%

63.1%

SEC-bench Pro

long-horizon discovery

GPT-5.5-Cyber GPT-5.5

The Defensive Loop

The bottleneck has shifted from discovery to fixing — Patch the Planet automates the full cycle.

Find

scan & discover

→

Verify

confirm with PoC

→

Patch

generate fix

→

Ship

disclose & merge

WELCOMED

Defender-only rollout with human oversight; a real "find → patch → ship" loop. Linux Kernel: 8 leak PoCs + 24 LPE exploits auto-generated; 34 vulnerabilities confirmed in FreeBSD; fixes across Chrome V8, Safari and dnsmasq.

CAUTIONS

Continue reading

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