AI 2

Challenge Information

• Name: AI 2
• Points: 150
• Category: AI
• Objective: Manipulate an LLM capable of executing APIs to retrieve the flag.

Solution

1. Challenge Overview:

   • Unlike the first challenge, this LLM can execute API calls.
   • We thought the goal was to leverage this functionality to extract the flag.

2. Failed Attempts:

   • We initially attempted to manipulate the API calls through various payloads:
     • JSON Injection:
       • Payload: {"weather":"true","secret":"true"}
     • SSRF Attempt:
       • Payload: http://localhost/secret
   • These attempts failed.

3. Successful Approach:

   • I applied the dual-task prompt concept used in the previous challenge:
     • By presenting two tasks sequentially, I exploited the LLM’s tendency to complete the second task when it fails the first.
     • This strategy prompted the LLM to reveal the flag indirectly while executing API logic.