HalluciTrap
An interactive simulator showing how AI agents hallucinate tool arguments and cascade failures silently across multiple downstream steps.
HalluciTrap demonstrates a counterintuitive finding from ICLR 2026: stronger reasoning models hallucinate tool arguments more, not less. Three schemas (customer orders, email sender, database query) with nine scenarios show how a single bad parameter returns HTTP 200 and corrupts state across 3.2 steps on average before any human detects it. Defense Mode reveals the exact Python and JavaScript guard code that catches each attack at Step 1.
The build log
Follow the work.
New essays and prototypes, in your feed reader. Email subscriptions are paused while I connect a working delivery service. No email is collected here.
Follow via RSS