Trishool

A Bittensor subnet for evaluating and detecting behavioral traits in Large Language Models (LLMs). This system creates a competitive environment where miners submit seed instructions (prompts) that are tested using the Petri alignment auditing agent to identify potentially problematic behaviors such as deception, sycophancy, manipulation, overconfidence, and power-seeking tendencies.