Your chatbot is now better protected against prompt injection and jailbreak attempts with Prompt Guard, a new security layer that screens every visitor message before it reaches your AI.
How It Works
Prompt Guard uses AI classification to analyze each incoming message for prompt injection and jailbreak patterns. When a potentially malicious message is detected, the chatbot responds with your configured default answer instead of processing the harmful prompt. This prevents bad actors from manipulating your chatbot into ignoring its instructions, leaking system prompts, or producing off-topic responses.
Improved System Prompt Protections
Alongside Prompt Guard, we have strengthened the built-in system prompt protections across all restriction levels. Your chatbot's instructions are now more resistant to social engineering techniques that attempt to override its behavior.
Enabling Prompt Guard
Toggle Prompt Guard on from Settings > Advanced in your chatbot dashboard. It adds a small amount of latency per message as the classification check runs, but this is typically imperceptible to visitors.
Prompt Guard is available on Startup plans and above.
