OpenAI's Daybreak embeds security into code from day one

OpenAI has unveiled Daybreak, an initiative built on its frontier LLMs and its Codex coding assistant to help developers build secure software from the ground up. Unveiled on 12 May 2026, Daybreak leverages the Trusted Access for Cyber program to reserve access to certain frontier models for a selective set of organisations, and includes GPT‑5.5 (general), GPT‑5.5 with TAC, GPT‑5.5‑Cyber, and Codex Security as a research preview.

While TAC focuses on vetted users identifying and fixing vulnerabilities, Daybreak aims to tackle security from the start of the software development lifecycle, with tasks such as codebase scanning, vulnerability triage and regression testing, and automated vulnerability detection and response.

As of May 2026, OpenAI said its TAC programme includes hundreds of organisations and thousands of defenders, including names such as Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, NVIDIA, Oracle, Palo Alto Networks, Sophos and Zscaler, alongside several large financial institutions. OpenAI also signalled plans to deploy cyber‑capable models in cooperation with industry and government partners, and to expand TAC to more government agencies.

Anthony Grieco of Cisco described frontier models as “powerful force multipliers for defenders,” emphasising the need for an enterprise‑ready framework to ensure trust and security.