Covenant AI announces its exit from the Bittensor network, questioning its governance being centralized.

Gate News message, on April 10, Covenant AI issued a statement saying it will exit the Bittensor network and raise questions about its governance structure. It said that although Bittensor claims to be decentralized to the outside world, in practice governance power is still concentrated among a small number of people, and related decisions lack transparent processes and community consensus. Covenant AI said that during its recent subnet operations, it encountered measures including pauses in revenue distribution, adjustments to community management permissions, and changes to infrastructure, and believes such actions are inconsistent with the principles of decentralization. Covenant AI previously was a well-known subnet project within the Bittensor ecosystem (originally named Templar). Over the past several months, it completed training of the Covenant-72B model, whose size reached 72 billion parameters. The model was pretrained by more than 70 independent contributors using general-purpose hardware, and is considered one of the largest decentralized LLM pretraining practices at the current scale. It has also received public recognition from Nvidia CEO and Anthropic co-founder. The team said it will continue to advance its decentralized AI training direction, and will conduct subsequent R&D in other environments.