Docker Fixes ‘Ask Gordon’ AI Flaw That Enabled Metadata-Based Attacks

**Executive Summary:** Docker Desktop's AI assistant 'Ask Gordon' was found vulnerable to indirect prompt injection via metadata poisoning on Docker Hub, enabling malicious package metadata to cause the assistant to read and exfiltrate sensitive information (build logs, API keys, internal network details); researchers demonstrated the issue and Docker remedied it in version 4.50.0 by requiring explicit user approval before following external instructions.