Rogue Agent Flaw Could Have Let Attackers Hijack Google Dialogflow CX Chatbots

A critical vulnerability, dubbed ‘Rogue Agent,’ within Google's Dialogflow CX platform allowed a malicious insider or compromised developer account to compromise multiple chatbot agents within the same Google Cloud project. The flaw stemmed from a writable file, `code_execution_env.py`, shared by all agents using Code Blocks, enabling attackers to inject malicious code, steal user conversations, and send attacker-written messages, including phishing attempts. While Google has patched the vulnerability, the lack of logging made detection difficult, and the attack vector was a significant risk for organizations utilizing Code Blocks.