Researcher Claims Control of ChatGPT Secure Sandbox

A researcher presented a proof-of-concept attack chain at Black Hat USA 2026 demonstrating C2-style influence over ChatGPT’s isolated code execution sandbox. The report frames this as gaining a degree of control over an environment meant to be isolated from the host and other sessions. No detail was given on patch status or the specific technique used. Sandbox escapes in AI coding/execution environments are a growing research area as these sandboxes see wider use in agentic workflows.