Lies-in-the-Loop Attack Turns AI Safety Dialogs into Remote Code Execution Attack

A Checkmarx report details a technique named "Lies-in-the-Loop" that weaponizes AI Human-in-the-Loop approval dialogs by injecting or padding malicious payloads so dangerous commands are hidden from users; when users approve seemingly benign dialogs, arbitrary code can run. The researchers demonstrated a proof-of-concept on multiple AI code assistants (including Claude Code and Microsoft Copilot Chat), highlighted risks when combined with Markdown injection to create fake dialogs, and warned this represents a fundamental challenge for AI agent design and HITL controls.