Show HN: Khaos – My Experiment Reveals All AI Agents Failed Within 30 Seconds

🚀 Introducing Khaos: The Open-Source Chaos Engineering Framework for AI Agents!

In a world where AI is increasingly handling sensitive data, we need robust testing methods. Khaos confronts this challenge head-on by adversarially testing AI agents designed for payment processing, database queries, and PII management. Here’s what you can expect:

• Challenge AI Safety: Identify vulnerabilities like prompt injection and data exfiltration.
• Hands-On Learning: Experiment with six intentionally vulnerable agents, including a payment processor and a support bot.
• Local Testing: Run Khaos 100% locally with no cloud dependency.

🔍 What Khaos Tests:

• Prompt Injection & Policy Bypasses
• Tool Misuse & Unauthorized Access
• Data Exfiltration Risks
• Fault Injection Scenarios

💡 Start testing today with the commands:

1. pip install khaos-agent
2. khaos discover
3. khaos run my-agent --pack security

Curious about the resilience of your AI agents? Dive into Khaos today! Explore the repo here! and share your insights! 💡🔗

Source link