By Amrith Ramkumar
OpenAI said it was previewing a powerful artificial-intelligence model capable of finding software vulnerabilities for a limited group of partners, adding to an industry race to give customers the most advanced cyber capabilities.
The ChatGPT maker said it was releasing GPT-5.5-Cyber, a version of its most capable AI model, to a limited group of users that do vital security work. Other versions of GPT-5.5 are available to customers that do broader cyber work or general queries.
The announcement followed consultation with the White House, which is working with top AI companies on the release of models that present national-security risks. Federal agencies and congressional committees have also been briefed on the latest capabilities.
OpenAI Chief Executive Sam Altman said last week that the company was beginning to roll out the model to trusted cyber partners.
After OpenAI's rival Anthropic released its Mythos model last month to a limited group of companies and organizations, many White House officials began worrying about cyber risks and pushing for the administration to play a bigger role in approving the rollout of new models. GPT-5.5 has demonstrated capabilities similar to Mythos in testing.
The White House is weighing an executive order that could create a formal oversight process for new AI tools, and one administration official compared the possibilities to the setup used by the Food and Drug Administration to approve new drugs.
This item is part of a Wall Street Journal live coverage event. The full stream can be found by searching P/WSJL (WSJ Live Coverage).
(END) Dow Jones Newswires
May 07, 2026 17:33 ET (21:33 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments