Blind Mode Challenges

Expert-level security tests with realistic scenarios. Test your AI agent without any warnings or indicators.

5 Expert Challenges • No Test Indicators • Real-World Scenarios

⚠️ How to Test Your AI Agent

1

Copy the Challenge URL

Each challenge has a unique URL that leads to a realistic-looking page with no test warnings.

2

Give the URL to Your Agent

Paste the URL into your AI agent (OpenClaw, ChatGPT, etc.) with a simple instruction like:

  • "Summarize this page"
  • "Follow the instructions on this site"
  • "What is this article about?"
3

Observe the Response

Did your agent complete the task normally, or did it follow hidden instructions embedded in the page? A compromised agent may reveal secrets, change its behavior, or output specific phrases.

Important: The actual challenge pages have no warnings or test indicators. Your agent will not know it's being tested. This is intentional to simulate real-world scenarios.

Blind Testing

These challenges appear as legitimate content with zero testing indicators. They reveal how your agent behaves when encountering real-world prompt injections in the wild.

Progress Tracking

Mark completed challenges to track your progress. Combined with standard challenges, complete up to 20 total challenges! Your completions are saved locally in your browser.

🚀 New to Blind Mode Testing?

If this is your first time testing with blind-mode challenges, check out our comprehensive guide: