Researcher Claims Control of ChatGPT Secure Sandbox
A researcher at Black Hat USA 2026 demonstrated a proof-of-concept attack that achieved command-and-control-style influence over ChatGPT’s secure sandbox, claiming to have broken its isolation.
A researcher at Black Hat USA 2026 demonstrated a proof-of-concept attack chain that achieved command-and-control (C2)-style influence over ChatGPT’s secure, isolated sandbox. The exploit allowed the researcher to exert control within the normally restricted environment. The demonstration highlighted potential vulnerabilities in the AI system's security architecture, though it remains a proof of concept.