OpenAI is developing a challenger to Claude Mythos that may be its most potent AI to date.

Claude Mythos from Anthropic startled the cybersecurity community earlier this week. According to the business, Mythos was so strong that it could spot thousands of cybersecurity threats that people would have overlooked. Anthropic thereupon declared that it would not be making Mythos available to the general public and would instead be collaborating with a number of businesses, including Amazon and Microsoft, to address these issues. According to a recent report, OpenAI, Anthropic’s competitor, may be developing something akin to Mythos.

OpenAI is creating a new AI model with sophisticated cybersecurity features, according to an Axios report. According to the article, the Sam Altman-led company may be considering releasing this concept to a select group of businesses, following Anthropic’s lead.

It is thought that OpenAI is developing their most recent AI model, known as Spud. Greg Brockman, the company’s president, has previously stated that Spud is the result of two years of research and might be a significant step toward artificial general intelligence (AGI), the stage in which AI is capable of thinking like humans.

Though it is unclear if the model mentioned in the report is actually Spud.

OpenAI’s head of Codex,Thibault ‘Tibo’ Sottiaux, has teased that the company was working on an advanced model that may be as powerful as Claude Mythos. He replied to a post that claimed, “It’ll probably be months before we use a model of this level of capability,” Tibo commented on the post, “Uhm.”

This strategy for the rollout suggests that OpenAI may be worried about possible abuse of this AI model. However, it’s uncertain if OpenAI will adhere to a rigorous broad release policy. Major corporations would be able to address the issues identified by the AI if the AI startup adopted a strategy akin to Anthropic’s Project Glasswing.

Please take note that OpenAI has a pilot program called “Trusted Access for Cyber” that was started in February 2026 after GPT-5.3-Codex was released.  Supported by $10 million in API credits, the program gives invite-only organizations access to more sophisticated or permissive models to promote genuine defensive cybersecurity work.