Breaking

OpenAI Rolls Out GPT-5.5-Cyber for Defenders and an Open-Source Patch Drive

June 22, 2026 at 23:58 EDT

Security
Foundation Models
Open Source

Around June 22, 2026, OpenAI expanded its cybersecurity initiative "Daybreak," launching a limited preview of a defender-focused model called "GPT-5.5-Cyber" and an open-source vulnerability-fixing effort, "Patch the Planet," run jointly with Trail of Bits, the company announced.

June 22, 2026 · OpenAI

Daybreak: AI Shifts the Security Fight from Finding Bugs to Fixing Them

OpenAI expands its cybersecurity effort with a full release of GPT-5.5-Cyber for trusted defenders and launches Patch the Planet — an initiative to fix vulnerabilities across 30+ open-source projects, from cURL and Python to the Linux Kernel.

85.6%

GPT-5.5-Cyber on the CyberGym benchmark

30M+

Linux Kernel lines scanned for security-relevant components

<1 day

To build a fuzzing lab — versus weeks done manually

Benchmark: CyberGym score

Higher is better · vulnerability-focused evaluation

81.8%

GPT-5.5

85.6%

GPT-5.5-Cyber

Patch the Planet · First-week results

Run with Trail of Bits, HackerOne and others across open-source projects.

projects

pull requests

issues filed

auto-generated kernel PoCs*

*8 pointer information-leak PoCs + 24 local privilege-escalation exploits, dynamically validated.

The new bottleneck: discovery is fast, remediation is scarce

AI finds a firehose of flaws

→

Humans triage & review

→

Patches reach merge

Why defenders welcome it

High-signal, practical output praised by Trail of Bits
Strong judgment in building fuzzing labs
Used for malware analysis, DDoS investigation, red teaming

Open concerns

Manual fixing may not keep up with findings
False positives demand human review
Limited access raises the bar to use

Continue reading

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