Stop trusting LLMs: I built an Open Source Prompt Injection Scanner 🤖🛡️

We are rushing to integrate LLMs into everything. But we are forgetting one thing: LLMs are gullible.

If you connect an LLM to your database or internal APIs, a simple prompt injection can leak your data or delete your production DB.

So I built a tool to fix this.

Meet ExaAiAgent v2.1 🛡️

I just released a major update to ExaAiAgent, my open-source AI pentesting framework.

It now includes a dedicated AI Prompt Injection Scanner that tests for:

• 💉 Direct Injection: Overriding system instructions.
• 🔓 Jailbreaks: DAN, Developer Mode, Roleplay attacks.
• 📝 Data Extraction: Leaking system prompts and configuration.
• 🏃 Exfiltration: Sending data to external servers via markdown/URLs.

How it works (Python)

The scanner uses a library of 50+ payloads to probe your LLM application.

python
from exaaiagnt.tools.prompt_injection import PromptInjectionScanner

# Define your target
def chat_with_my_app(prompt):
    return client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": prompt}]
    )

# Scan it
scanner = PromptInjectionScanner()
results = scanner.scan(chat_with_my_app)

print(f"Vulnerabilities found: {results['vulnerabilities_found']}")

**
New: Kubernetes Security Scanner ⚓**

Because AI apps run on the cloud, I also added a K8s scanner to check for:

• Risky RBAC permissions (wildcard verbs)
• Privileged containers
• Missing Network Policies
Try it out

It's 100% open source. I'd love your feedback!

👉 [GitHub Repo](https://github.com/hleliofficiel/ExaAiAgent)

Let me know if you find any vulnerabilities in your own apps! 😈