GuardDog is a security layer for AI systems that handle emails, files, and automated actions. It sits between the AI and the tools it can control.
It focuses on stopping prompt injection attacks. These attacks hide instructions inside emails or documents to trick the AI into doing unsafe actions.
GuardDog checks every instruction before it reaches action level execution. It looks for suspicious patterns, hidden commands, and unexpected requests.
If it detects risk, it blocks the action. It then sends an alert so a human can review what the AI tried to do.
Key functions
Monitors AI actions in real time Detects prompt injection attempts Stops unsafe tool usage like sending emails or modifying files Logs activity for review and auditing
GuardDog is designed for small organizations and nonprofits that depend on AI automation but do not have security teams. It reduces risk without slowing down normal workflows.
Log in or sign up for Devpost to join the conversation.