Covenant AI, a subnet developer that has been quite active on Bittensor, recently revealed that it has decided to leave the AI-powered network ecosystem. As mentioned in a post on social media platform X, Covenant AI Founder Sam Dare claimed that the company may no longer develop anything on Bittensor since the web3 network’s governance mechanisms actually go against its stated objective to maintain decentralization.
Dare stated via X that the main objective of Bittensor, the actual promise that drew web3 builders, crypto miners, blockchain / DLT network validators, and investors into this ecosystem, is that no single entity would be able to control it.
— covenant (@covenant_ai) April 9, 2026
Like many other crypto industry participants have claimed about most other projects, this so-called promise made by Bittensor is a lie (according to statements from Dare).
The Covenant founder also mentioned that the team has been developing on a conviction or firm belief that AI model training must not be controlled by a single, centralized entity.
They have been sharply focused on delivering on that objective by implementing Covenant-72B into one of the largest decentralized instances of LLM pre-training run in history.
These ongoing efforts had heleped Covenant become one of the most popular subnets on Bittensor, the team claimed.
However, Covenant has now made the decision to exit the network, while claiming that Bittensor Co-founder Jacob Steeves’ actions have been the primary motivation for this.
Dare added that Steeves, also referred to as Const, has been imposing his authority over the subnet in order to regain control over Covenant after it had become too large to be properly managed.
During the past few years, Bittensor has reportedly functioned / operated under a so-called “triumvirate” structure with three different people managing the multisig for handling any network updates, touting it as distributed governance to the ecosystem participants.
Dare further claims that this is not actually the case and then refers to it as decentralization “theatre.”
They added that Jacob Steeves actually maintains effective control over the triumvirate, resists meaningful transfer of authority, and even “deploys changes unilaterally whenever he chooses, without process and without consensus.”
Dare pointed out that Steeves has taken certain measures against Covenant that seem to compromise Bittensor’s decentralization ethos / principles.
These measures reportedly include actively suspending any emissions to Covenant’s subnets.
They also include getting rid of the team’s moderation capabilities over its community channels, as well as deprecating its subnet infrastructure.
Furthermore, Dare has accused Steeves of applying so-called economic pressure tactics via fairly large, visible digital token sales that were reportedly timed to along with operational conflicts.
According to Dare, they were “punitive actions” that were intentionally taken by one individual who never actually gave up control of a network he “tells the world he no longer runs.”
For added background and context, Bittensor (TAO) is described as an open-source, “decentralized” blockchain protocol that aims to create a P2P marketplace for machine learning models.
According to its developers, it enables AI algorithms to be produced in a collaborative manner.
It is rewarding or incentivizing participants with TAO crypto tokens for contributing intelligence to a “shared, global network, offering an alternative to corporate-controlled AI.”
However, this project is not the only one that claims to be decentralized but really isn’t. Perhaps web3 founders think that promoting their projects in this manner will help with their image and lead to greater adoption. But in the long-term, it actually makes things a lot worse. That’s why it is better to deliver on stated goals and promises instead of making tall claims that never materialize.