Inspiration

As AI agents become increasingly autonomous, a single unauthorized action can trigger multiple dependent operations. We wanted to address this security gap by creating a system that verifies actions before execution and prevents risks from spreading across workflows.

What it does

ChainShield AI is a dependency-aware security verification layer for autonomous AI workflows. It checks proposed actions against configurable authorization policies, identifies workflow dependencies, and blocks unauthorized actions along with dependent operations. It also provides workflow visualization, verification results, and audit logs.

How we built it

We built ChainShield AI using Python and Streamlit. NetworkX handles workflow dependency modeling, while Pandas supports structured data handling. The prototype demonstrates policy-based verification, dependency-aware blocking, and audit logging through simulated workflows, including a refund authorization scenario.

Challenges we ran into

One major challenge was identifying how an unauthorized action could affect subsequent dependent operations. We also worked on designing configurable security policies, preventing unnecessary blocking of legitimate actions, and presenting complex security decisions through an understandable interface

Accomplishments that we're proud of

We are proud to have transformed our idea into a working prototype that brings policy-based verification, dependency analysis, workflow visualization, and audit logging together in one interactive dashboard.

What we learned

We gained practical experience in Python, Streamlit, NetworkX, workflow modeling, authorization policies, and collaborative development. We also learned why security verification and dependency analysis are essential for trustworthy autonomous AI systems

What's next for Example

We plan to integrate ChainShield AI with real AI agent frameworks, introduce human-in-the-loop approvals, improve risk scoring, expand security testing, and develop it into a model-independent security layer for autonomous workflows.