Demo: Secure generative AI apps | Akamai Firewall for AI

Akamai Firewall for AI Demo

Ziv Eli demonstrates Akamai Firewall for AI protecting a Cloudy Airlines chatbot from prompt-injection attacks.

The interface allows operators to enable or disable the firewall and choose how to handle detections: monitor, deny, or modify. A live console displays chatbot activity, detections, and actions taken. The demo also reveals the chatbot’s system prompt, which instructs it not to disclose a coupon code.

Demo Results

In monitor mode, repeated prompt-injection attempts were detected but not blocked. The chatbot responded differently to the same attack: one response revealed part of its system prompt, while another also disclosed the coupon code. This demonstrates the risk of unpredictable chatbot outputs.

In deny mode, the same attack was blocked. The chatbot reported that the conversation violated policy, and the console recorded both the prompt-injection detection and the blocked-input action.

Main Speaker

• Ziv Eli, presenting the Firewall for AI demo