OpenAI ships GPT-5.5-Cyber and 'Patch the Planet' for open-source security

Daybreak's June 22 expansion bundles a cyber-tuned frontier model, a Codex security plugin, and a Trail of Bits partnership to fix bugs in critical open-source software at scale.

OpenAI on June 22 graduated its Daybreak cybersecurity program out of preview, shipping the full version of GPT-5.5-Cyber alongside three new pieces: an updated Codex Security plugin for development toolchains, a partner program that lets vendors embed GPT-5.5 in their own products, and an open-source initiative called Patch the Planet. OpenAI puts GPT-5.5-Cyber at 85.6% on the CyberGym vulnerability-reproduction benchmark, up from 81.8% for the standard GPT-5.5. The partner program launches with Accenture, Cisco, CrowdStrike, IBM, Okta, Palo Alto Networks, and Wiz integrating GPT-5.5 under a Trusted Access tier.

Patch the Planet is the most interesting piece. Built with Trail of Bits and HackerOne, it pairs AI-assisted security research using GPT-5.5-Cyber with human review by Trail of Bits engineers before any finding reaches a maintainer. Initial participants include cURL, the Go project, Python, Sigstore, pyca/cryptography, NATS, aiohttp, freenginx, and python.org — the connective tissue under most of the internet. In its first week, OpenAI says the program produced 64 pull requests and 51 issues across 19 projects, with 37 patches already merged.

The framing matters. For roughly a decade the bottleneck in software security was finding vulnerabilities; the patching pipeline was slow but tractable. LLMs flipped that. Models can now surface candidate bugs at a rate human maintainers cannot triage, which is why the new generation of security tooling is going after the patch step. Anthropic ran a similar shape with Project Glasswing earlier this year. The shipping date is also telling — June 22 was the same day the Five Eyes intelligence agencies issued a rare joint warning that AI-powered offensive cyber operations are months away, not years.

Takeaway for learners: AI-assisted vulnerability work is becoming a real specialty, and the productive loop is not 'let the model patch it.' It is 'let the model surface candidates, let a human verify and shape the fix, let the maintainer ship.' That triage layer in the middle is where most of the value lives, and it is a job a curious early-career engineer can learn into. If you have ever wanted to contribute to open source, walking into a Patch the Planet project as a reviewer of GPT-5.5-Cyber's outputs is one of the most leveraged things you can do with an afternoon.