ChatGPT Sandbox C2 Attack Demonstrated at Black Hat 2026
A researcher at Black Hat USA 2026 demonstrated a proof-of-concept attack chain enabling command-and-control-style influence over ChatGPT's isolated execution sandbox. The technique represents a significant escalation in LLM exploit sophistication, moving beyond prompt manipulation toward infrastructure-level session control. If reproducible at scale, this class of attack could undermine the isolation guarantees that underpin safe AI code execution environments.