On Monday, OpenAI announced a project called “Daybreak,” which CEO Sam Altman says aims to “accelerate cyber defense and continually secure software.”
OpenAI is launching Daybreak, our effort to accelerate cyber defense and continuously secure software.
AI is already good and is about to get great at cyber security; We would now like to start working with as many companies as possible to help them continue to protect themselves.
– Sam Altman (@sama) 11 May 2026
The OpenAI blog post announcing Daybreak doesn’t mention the word “project” at all, perhaps making it a little less apt for readers to compare it to Anthropic’s Project Glasswing, but check it out: It sounds very similar to Anthropic’s Project Glasswing. Like Project Glasswing, this is a program in which a leading AI company seeks to partner with corporate and government entities to root out security vulnerabilities using OpenAI’s most advanced models, in the hopes of “seeing risk earlier, taking action sooner, and helping software be resilient by design.”
Glasswing was launched last month with Anthropic’s announcement of its Cloud Mythos preview model, a model so capable – at least according to its creators – that it threatened the world. As Anthropic’s system card for the model explained:
Due to the significant growth in capabilities of Cloud Mythos Preview, we have decided not to make it generally available. Instead, we are using it as part of a defensive cybersecurity program with a limited number of partners.
In other words, because this is the “most cyber-capable model” that Anthropic has ever made, it should be discontinued for now, unless you’re a VIP. Influential software developer Daniel Steinberg called it “certainly a surprisingly successful marketing stunt”.
Two days after that announcement, reports began to emerge about a similar project at OpenAI. An anonymously sourced Axios story described it as “a product with advanced cybersecurity capabilities that it plans to release to a small group of partners.”
The Daybreak announcement is much more public-facing than that, and significantly less ominous and secretive than Project Glasswing. There are two buttons at the top of the page: “Request a Vulnerability Scan” and “Contact Sales.” When you click “Request a Vulnerability Scan” you get a short and non-challenging form:
Altman said in his X post that OpenAI would like to “start working with as many companies as possible now,” and to be fair, that’s how the effort is accomplished. This is refreshing compared to the way Project Glasswing was launched, with frightened governments scurrying around behind the scenes like agitated ants.
The announcement says Daybreak uses Codex Security, which was announced as a research preview in March, to build a “threat model” of a given system, which outlines its functions, what the system trusts, and therefore what the vulnerabilities are. As a reference, it digs into your actual codebase for real-world exploits.
Then, in theory, this Daybreak patches them.
<a href