AgentShield AI Security Scanner

Inspiration

What it does

How we built it## Inspiration

As AI agents become widely used, unauthorized access and malicious prompt attacks bring serious security risks. We hope to build an automatic detection tool to find vulnerabilities in AI agents and protect LLM applications from misuse.

What we learned

We learned the common types of prompt injection and agent permission loopholes. We also mastered how to construct adversarial test prompts and analyze the response results of large language models automatically.

How we built it

We built the backend with Python and integrated LLM APIs to run batches of security tests. We developed a simple web UI to display scanning reports, and optimized the testing queue to improve scanning efficiency. Finally we deployed the service as a hackathon prototype.

Challenges we faced

The main challenge was filtering false positive warnings. We solved this by adding multi-round result verification. We also struggled with limited API call quotas, so we optimized test cases to reduce unnecessary requests.

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AgentShield AI Security Scanner

Built With

• api