Odaily Planet Daily reports that Covenant AI has issued a statement announcing its withdrawal from the Bittensor network and raising concerns about its governance structure. It stated that, despite Bittensor's public claims of decentralization, actual governance power remains concentrated in the hands of a few, with decision-making lacking transparent processes and community consensus.
Covenant AI stated that recent measures during the operation of its subnet—including the suspension of revenue distribution, adjustments to community governance rights, and infrastructure changes—contradict the principles of decentralization. The team affirmed its commitment to advancing decentralized AI training and will continue subsequent research, development, and project deployment in other environments.
Covenant AI, formerly known as Templar, is a well-known subnet project within the Bittensor ecosystem. Over the past several months, the team completed training of the Covenant-72B model, a 72-billion-parameter model pre-trained by more than 70 independent contributors using commodity hardware. It is regarded as one of the largest decentralized LLM pre-training efforts to date and has received public recognition from NVIDIA’s CEO and the co-founder of Anthropic. The team plans to continue advancing decentralized AI training and will pursue further development in additional environments.