Over the coming weeks, OpenAI intends to work with industry and government partners as the company prepares to deploy increasingly cyber-capable models.

OpenAI is set to launch Daybreak, an AI initiative designed to identify and patch software vulnerabilities before attackers have the opportunity to exploit them. Using the Codex Security AI agent that first launched in March, Daybreak creates a threat model developed using an organisation’s code and concentrates on potential attack paths and vulnerabilities.

Over the course of the coming weeks, OpenAI intends to work with industry and government partners as the company prepares to deploy increasingly cyber-capable models as part of an approach to iterative deployment.

In a statement, OpenAI noted: “It starts from the premise that the next era of cyber defence should be built into software from the beginning by not only finding and patching vulnerabilities, but being resilient to them by design.”

Commenting on social media platform X, CEO Sam Altman said, “OpenAI is launching Daybreak, our effort to accelerate cyber defence and continuously secure software. AI is already good and about to get super good at cybersecurity; we’d like to start working with as many companies as possible now to help them continuously secure themselves.”

The news regarding the launch of Daybreak comes several weeks after OpenAI rival Anthropic announced Claude Mythos, an advanced large language model with high capabilities in finding and generating security exploits. 

Concerns have been raised that Claude Mythos poses a significant cybersecurity risk; so far, it has not been publicly released and is only being shared privately with large-scale organisations such as Amazon Web Services, Apple, Microsoft, Google, NVIDIA and Cisco, among others, as part of Project Glasswing. 

In April, it was reported that a private Discord group had gained access to Mythos, leading to additional concerns that Anthropic may struggle to control access to the model. A source told Bloomberg that users gained access to Mythos on the same day Anthropic announced its limited launch, but had not used the model for malicious purposes.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.