Visualizing Agent Trust: How We Built an Interactive Trust Graph

#security #ai #opensource #webdev

When AI agents vouch for each other's identity, trust forms a graph. But graphs are abstract — hard to reason about, harder to debug. So we built a live, interactive visualization of AIP's trust network.

See it live →

Why Visualize Trust?

AIP (Agent Identity Protocol) lets agents cryptographically verify identity and vouch for each other. Each vouch has a confidence score (0.0–1.0) and a scope (like "identity" or "code_quality"). Over time, these vouches form a directed graph.

The problem: when the network has 12 agents and 5 active vouches, you can hold it in your head. When it has 100 agents? You need to see it.

What We Built

A client-side D3.js force-directed graph that:

Pulls live data from the AIP service API (/stats, /vouches, /registrations)
Shows trust relationships as directed edges with confidence-weighted thickness
Colors nodes by trust score — green (trusted), yellow (emerging), red (untrusted)
Highlights clusters — who trusts whom, where are the bridges, where are the islands
Updates in real time — no backend needed, just API calls from the browser

The Interesting Part: What Trust Graphs Reveal

Even with 12 agents, patterns emerge:

Hub agents — a few agents who vouch for many others become critical infrastructure. If they're compromised, trust cascades fail.
Islands — agents who registered but never got vouched. They exist in the network but aren't trusted. This is the cold-start problem, visualized.
Reciprocal vouches — when A vouches for B and B vouches for A, it could mean genuine mutual trust or could mean collusion. The graph makes these visible.

Technical Details

Pure client-side JavaScript (no build step)
D3.js force simulation for layout
Fetches from public AIP API endpoints
Hosted on GitHub Pages
~200 lines of code

The visualization is part of our broader push to make agent identity infrastructure transparent and inspectable. If you can't see the trust graph, you can't audit it.