OpenAI launches Daybreak to find and fix software security flaws

In short: OpenAI has launched Daybreak, a new security program that uses AI to help organizations find and fix software weaknesses before attackers do.

What happened

OpenAI introduced Daybreak, an AI initiative focused on finding vulnerabilities, meaning weak spots in software that hackers can break into. The goal is to catch these problems early and help teams patch them.

Daybreak uses OpenAI’s Codex Security AI agent, which OpenAI launched in March. OpenAI says the tool can look at an organization’s code, map out how an attacker might try to get in, and check whether certain weaknesses are likely real. It then automates the search for the higher risk issues, so security teams can focus on the most urgent fixes.

OpenAI says Daybreak is not based on a single model. Instead, it combines several OpenAI systems, including Codex, specialized cyber-focused models like GPT-5.5-Cyber, and work with security partners.

The launch comes a little over a month after Anthropic announced Claude Mythos, another security-focused AI model. Anthropic said it was too dangerous to release publicly and only shared it privately through a program called Project Glasswing. The Verge reports that despite that approach, some unauthorized parties still got access.

Why it matters

Security problems in software can lead to account takeovers, stolen data, and service outages. Tools like Daybreak are meant to act like an extra set of eyes, similar to having a tire inspector check for cracks before a long road trip (small issues can become big problems later). If these systems work as promised, they could help organizations fix more issues faster, but they also raise questions about how widely powerful security tools should be shared.

Source: The Verge AI