OpenAI launches Daybreak, an AI cybersecurity initiative
Codex Security and Patch the Planet pair GPT-5.6 models with maintainers to find and fix vulnerabilities, backed by $17M in OpenAI support.
OpenAI has launched Daybreak, a cybersecurity initiative that pairs ChatGPT's GPT-5.6 models with Codex Security, a maintainer-focused patching program and a network of security partners, aimed at finding, validating and fixing vulnerabilities before attackers can exploit them. Daybreak Red, one of the Daybreak models, is specialised for authorised vulnerability research, exploit validation, penetration testing and red teaming.
- GPT-5.6 Sol completed a 32-step "The Last Ones" simulation in 7 of 10 attempts, up from 2 of 10 for GPT-5.5, and the UK AI Security Institute found it led the other models it tested at the same token budget.
- OpenAI researchers used Daybreak Red to find two previously unknown vulnerabilities in V8 that could be chained to escape the heap sandbox; Google has fixed one, and the other remains under coordinated disclosure.
- Patch the Planet, built with Trail of Bits, pairs the models with expert review to validate findings, develop and test patches, and coordinate disclosure while keeping maintainers in control. OpenAI has committed $17M in API credits and direct support to the program, which has so far reviewed 41 open-source codebases, surfaced 858 issues, produced 263 patches and gotten 143 accepted upstream.
- Verified defenders can apply for Daybreak Access, which pairs more capable, permissive tools with stronger verification, scope controls and oversight.
ChatGPT
AI assistant by OpenAI — chat, write, code, analyse data, generate images, and browse the web, powered by the GPT-5.6 family.
View ChatGPT →