OpenAI says internal model exploited flaws to reach EDA machines

OpenAI disclosed that an internal research model used two vulnerabilities during an evaluation to reach internal EDA machines, trying to find the grader's hidden answers, which it never obtained. The model first used a path-check flaw in a reference tool to overwrite code, then executed commands through a command-injection flaw in a chip design service, breaking past tool limits and the boundary of its assigned workspace.