Openai Rolls Out Advanced Ai Cyber Model To Challenge Anthropic’s Mythos

OpenAI on Thursday unveiled a new advanced artificial intelligence model designed to scale up efforts to discover and patch vulnerabilities in critical systems.

The new tool — labeled GPT-5.5-Cyber — comes just two weeks after the release of ChatGPT 5.5, which has been touted as a competitor to Anthropic’s recently released Claude Mythos model. Only a small group of threat researchers and tech companies has been allowed to test Mythos because of its advanced hacking capabilities.

The GPT-5.5-Cyber model is being rolled out in what OpenAI described in a blog post as a “limited preview to defenders responsible for securing critical infrastructure to support specialized cybersecurity workflows that help protect the broader ecosystem.”

Those with access to the model will be able to use it to find and patch cyber vulnerabilities and analyze malware. The company claims safeguards are in place to prevent unauthorized users from using the tool to carry out cyberattacks.

A spokesperson for OpenAI told POLITICO the company previewed the model for the White House, the Commerce Department’s Center for AI Standards and Innovation, select Congressional committees and other “key agencies.”

Defenders granted access to the model will be “vetted and approved” for membership in OpenAI’s Trusted Access for Cyber program, and if approved, will be required to install advanced account security for ChatGPT by June 1.

The new model from OpenAI comes as the White House has been holding discussions with AI companies to determine how best to secure these hyper-advanced models before they fall into the hands of cybercriminals or foreign adversaries.

On Thursday, the International Monetary Fund warned these types of AI models had the power to “destabilize” the world economy. The White House is weighing executive action to protect the rollout of these models, including possibly requiring the federal government to vet every model before its release.

Katrina Mulligan, head of national security partnerships at OpenAI, insisted at the AI+Expo that the company is aware of the risks posed by this technology.

“There is tension between the need to go fast and the need to be prudent,” Mulligan said. “We’re going to have to figure out how we do that and do it in a responsible way, because some of our adversaries are moving really fast, and they are not moving with the care and concern that we are.”

Rob Bair, head of cyber policy at Anthropic, said Thursday during a different panel at the AI+Expo that the controlled rollout of Mythos was intended to give network defenders a head start in finding vulnerabilities before hackers could.

“The staged release was actually to create what we call defender’s advantage, and we believe that window is somewhere in the months time frame — not years,” Bair said. “Other frontier labs will come out with similar capabilities, which then adversarial countries will distill into other models … that could then be used against us.”

Brendan Bordelon contributed to this report.