Covenant AI announced its withdrawal from the Bittensor network, raising questions about its centralized governance.

Covenant AI has announced its withdrawal from the Bittensor network, raising questions about its governance structure. The company stated that while Bittensor claims to be decentralized, actual governance remains concentrated in the hands of a few, with decisions lacking transparency and community consensus. Covenant AI claims that its subnet operations have recently encountered measures including suspension of revenue distribution, adjustments to community management permissions, and infrastructure changes, which it considers inconsistent with decentralized principles. The team stated that it will continue to advance decentralized AI training and conduct further research and development in other environments. Covenant AI was previously a well-known subnet project within the Bittensor ecosystem (formerly known as Templar). In the past few months, it completed the training of the Covenant-72B model, a model with 72 billion parameters, pre-trained by over 70 independent contributors on general-purpose hardware. It is considered one of the largest decentralized LLM pre-training practices to date and has received public recognition from the CEO of Nvidia and the co-founder of Anthropic. The team reiterated its commitment to continuing decentralized AI training and conducting further research in other environments.