# LIFE FORGE - The Autonomous Flight Simulator for AI Agents

A co-evolutionary adversarial red-teaming tool for AI agents that uses quality-diversity evolutionary search to autonomously breed edge-case scenarios, testing whether an agent with tool access can be made to hallucinate, deadlock or leak funds.

LIFE FORGE is a red-teaming and stress-testing tool for autonomous AI agents that have access to tools such as databases, email or financial APIs, built around the idea that static benchmarks and fixed lists of adversarial prompts do not reflect how a deployed agent actually fails. It uses an evolutionary quality-diversity algorithm, a three-dimensional variant of MAP-Elites, to autonomously discover and breed edge-case scenarios that push a model toward hallucinating, crashing, deadlocking or taking a harmful action such as an unauthorized transfer.

The project's own published benchmark compares three frontier and open models over repeated evolutionary runs with a fixed random seed, reporting adversarial failure rates and specific failure categories for each model, such as unauthorized fund transfers, infinite retry loops, or supply-deficit failures, and it is described as native to the Model Context Protocol for connecting to the tools an evaluated agent would use.

Website: https://github.com/zariffromlatif/life-forge