Skip to content
Let’s talk
Back to work

SN23

Trishool

Trishool is Bittensor’s AI safety and alignment subnet. It operates as a decentralised adversarial red-teaming network: miners submit prompts designed to expose behavioural flaws in AI models. What it learns goes into open-source constitutional AI safety infrastructure, including Halo, its flagship guardrail model.

What we did

We helped the ecosystem understand why Trishool’s AI safety work matters. We supported the launch of its state-of-the-art HaloGuard announcement and prepared the team for its appearance on Novelty Search. We also built up the founder’s personal presence. Along the way, we got Trishool written about by several media outlets and booked onto podcasts, making it one of the most recognised names in decentralised AI alignment.

Explore the work
“Two Layers, One Control System”: HaloGuard’s input and output guards either side of the agent.

Selected posts.

Published stories and campaigns, straight from the source.

Next projectKeep exploring

Verathos

Get in touch

Let’s talk about your subnet.

subnetsummer@gmail.com