Trishool

About Trishool

Trishool is an AI alignment and security protocol built on Bittensor, designed to protect advanced AI systems from adversarial attacks such as jailbreaks, prompt injections, and agent hijacking. It operates as a collaborative, miner-driven adversarial training network (subnet 23), where participants submit attack payloads against guarded AI agents. Successful exploits are used to retrain and harden the guard models in real time, with a new defense cycle every 7 days. The protocol features decentralized scoring, on-chain verification, and incentivizes the discovery of vulnerabilities by rewarding contributors in TAO. Trishool aims to ensure AI systems remain aligned with human values and resilient against evolving threats.