What is Patch the Planet? OpenAI + Trail of Bits Initiative (Jun 23, 2026)
What is Patch the Planet? OpenAI’s Open-Source Security Initiative
Patch the Planet is OpenAI’s open-source security program, launched June 23, 2026 alongside the full release of GPT-5.5-Cyber and the Codex Security IDE plugin. It is a direct strategic mirror of Anthropic’s Project Glasswing. Here’s what it does, what shipped, and why it matters.
Last verified: June 23, 2026.
The one-paragraph answer
Patch the Planet uses OpenAI’s restricted-access cybersecurity model (GPT-5.5-Cyber) plus Codex agent automations to systematically scan, validate, and patch vulnerabilities in critical open-source software. The work is coordinated with Trail of Bits (security partner) and HackerOne (disclosure partner). The first five-day sprint surfaced hundreds of issues across 30+ projects, merged dozens of patches, and produced reusable testing workflows other projects can adopt. The political-economic point of the program — not hidden in OpenAI’s framing — is to make the Daybreak platform “too important to shut down” in the same way Glasswing protected Anthropic during Fable 5 suspension.
What launched today
| Component | What it is |
|---|---|
| GPT-5.5-Cyber (full) | Cybersecurity-tuned model, 85.6% on CyberGym (vs 81.8% standard GPT-5.5), restricted to verified defenders |
| Codex Security plugin | IDE plugin embedding vulnerability scanning into the developer workflow |
| Patch the Planet | Operational program running Codex /goal sweeps on critical OSS, partnered with Trail of Bits + HackerOne |
The three pieces together are OpenAI’s complete cybersecurity stack response to Anthropic Glasswing.
What the first sprint actually produced
OpenAI ran a five-day sprint across multiple participating projects in the lead-up to launch. Reported outcomes:
- Hundreds of security issues surfaced across the cohort
- Dozens of patches merged into upstream projects
- Reusable testing workflows published, including fuzzing, variant analysis, and differential testing harnesses
- Trail of Bits Linux kernel demo: GPT-5.5-Cyber identified security-relevant components across 30M+ lines of code, validated issues dynamically, generated 8 kernel pointer information leak proof-of-concepts and 24 local privilege escalation exploits
The Linux kernel result is the headline capability demo. It would normally take a top-tier security team weeks to produce — and the model produced it in a single Codex /goal sweep with human review.
The initial cohort
Five projects publicly confirmed for the first wave:
- cURL — Daniel Stenberg’s data transfer library, embedded in essentially every connected device
- Go — Google’s programming language and standard library
- Python (CPython) — language runtime
- Sigstore — artifact signing for supply chain security
- pyca/cryptography — Python cryptographic primitives
OpenAI says 30+ total projects have committed. Trail of Bits and HackerOne handle operational coordination.
Why critical OSS, not enterprise?
Because OSS fixes are the highest-leverage way to demonstrate civic value. Critical OSS underlies the entire enterprise stack — every fix in cURL ripples to billions of devices, every fix in CPython ripples to nearly every cloud workload. Fixing OSS:
- Generates measurable, audit-friendly outcomes (CVEs filed, patches merged)
- Builds trust with government and standards bodies
- Creates a political case for keeping the underlying model accessible during future restrictive policy actions
This is exactly what Glasswing achieved for Anthropic. It is exactly what Patch the Planet is designed to achieve for OpenAI.
How it differs from Glasswing
| Dimension | Patch the Planet | Project Glasswing |
|---|---|---|
| Lab | OpenAI | Anthropic |
| Model | GPT-5.5-Cyber (full) | Mythos / Glasswing-tuned variants |
| Operational partners | Trail of Bits, HackerOne | 50+ partners across infra, OSS, government |
| IDE surface | Codex Security plugin (new today) | Claude Code defensive workflows |
| EU posture | Daybreak platform — no formal EU anchor yet | ENISA cooperation announced June 18, 2026 |
| Proof point | First Linux kernel demo, June 23, 2026 | Survived Fable 5 US suspension |
Glasswing has more partner depth and an EU government anchor. Patch the Planet has a more developer-embedded surface via Codex Security.
What this means for OSS maintainers
If you maintain a critical open-source project, three practical implications:
- You may be approached. Trail of Bits and HackerOne will be coordinating expansion of the cohort. If your project is widely deployed in critical infrastructure, expect outreach.
- You can adopt the workflows without joining. The fuzzing harnesses, variant analysis configs, and differential testing setups OpenAI is publishing are usable independently. The model access is restricted; the testing playbooks are public.
- HackerOne disclosure hygiene matters more. With AI-driven scans being run by labs, disclosure quality becomes the bottleneck. Make sure your project’s security policy and HackerOne (or equivalent) integration are clean.
The unspoken third question
The third question every OSS maintainer should ask: what happens when the same model that finds bugs for OpenAI’s authorized defenders gets stolen, leaked, or replicated? GPT-5.5-Cyber’s offensive capability is real — 24 local privilege escalations from a single Codex sweep on the Linux kernel is not a defensive-only capability. Trusted Access for Cyber is the human-review wrapper that keeps that asymmetry on the right side. If that wrapper fails, the same engine becomes a vulnerability-discovery weapon. The political case for restricted access is that this asymmetry is too dangerous to commoditize — and the OSS contribution program is the credibility that buys the restricted access.
Sources
- OpenAI Daybreak / Patch the Planet launch coverage, June 23, 2026
- AIToolsRecap, “AI News June 23 2026,” for sprint outcome reporting
- Trail of Bits Linux kernel demo write-ups
- CyberGym benchmark leaderboards (llm-stats.com, benchlm.ai)
Verified June 23, 2026.