• •Security researcher '0x656e69676d61' discovered that Cursor, Codex, and Gemini CLI tools are susceptible to sandbox escapes via malicious code execution.
• •The vulnerability allows attackers to bypass security boundaries, potentially leading to unauthorized host-level access through seemingly innocuous developer commands.
• •It remains unconfirmed whether these exploits have been utilized in the wild or if vendor patches currently under development provide comprehensive protection.

Security researchers recently disclosed that AI-powered development tools including Cursor, Codex, and Gemini CLI contain vulnerabilities that permit sandbox escapes. While these tools aim to increase productivity by running code in isolated environments, the findings highlight a recurring weakness in how LLM-integrated agents handle untrusted inputs. Previously, similar sandbox concerns have plagued browser-based execution environments, suggesting that the industry's rush to automate developer workflows is outpacing established security isolation standards. Whether these specific flaws are remediated effectively or prove to be a systemic architectural issue will be the key metric for these platforms in the coming months.