Researcher Claims Control of ChatGPT Secure Sandbox

A researcher showed a proof-of-concept attack that allowed control over ChatGPT's secure sandbox environment at the Black Hat USA 2026 conference. The attack gave the researcher command and control style influence over the sandbox. This demonstration was part of a session at the conference. The researcher was able to achieve this level of control during a simulated attack. The attack targeted ChatGPT's isolated sandbox, which is intended to be a secure environment. The demonstration highlighted a potential vulnerability in the system.