TrishoolSN23 Price

Trishool (SN23) is a Bittensor-powered subnet focused on decentralized AI alignment and safety. It operates as a collaborative, adversarial network where global participants continuously red-team and stress-test AI models to uncover misalignment, deception, jailbreaks, and harmful behaviors.Core mechanics involve validators issuing periodic alignment challenges targeting specific risks (e.g., deception, sycophancy, CBRN, crime facilitation, or agent boundary violations).

Miners respond by submitting seed prompts, Python agents, or adversarial strategies that attempt to elicit unsafe outputs. Submissions are executed in sandboxed environments using tools like adapted versions of Anthropic’s Petri, scored by judge models, and verified on-chain. Top-performing discoveries are rewarded in TAO and integrated into a shared “Guardian” core model, with regular RLVR retraining to evolve the safety layer. The system creates a self-improving loop (AlignNet) where every exploit strengthens defenses. It also develops production tools such as Halo, a guard model for enterprise AI deployments, and has secured partnerships including the Claude Partner Network and Google for Startups Web3 Program.