OpenAI plans to give limited early access to a new model with advanced cybersecurity capabilities to only a small number of companies, Axios reported on April 9, citing a source familiar with internal discussions.
OpenAI’s move is similar to how Anthropic released Mythos Preview on a limited basis.
OpenAI unveiled an invitation-only “Trusted Access for Cyber” pilot programme after the February launch of GPT-5.3-Codex. At the time, OpenAI said it would provide participating companies with model access and $10 million in API credits for use in defensive security work.
Former government officials and security experts have warned over the past year that if AI models fall into the hands of malicious attackers, they could autonomously cripple water facilities, power grids and financial systems. The concern appears to be becoming reality.
Rob T. Lee (롭 리), chief AI officer at SANS Institute, said, “The capability for AI models to analyse code or detect vulnerabilities in legacy codebases already exists. It cannot be stopped.”
Adam Meyers, a vice president at CrowdStrike, said, “The capabilities Mythos shows are an alarm bell for the entire industry.”
Stanislav Fort (스타니슬라프 포트), chief executive of security company AISLE, said, “If you are more concerned about the capability to develop new exploits than detecting model vulnerabilities, it is reasonable to disclose it in stages.”
Click Here For The Original Source
