Covenant AI Exits Bittensor, TAO Drops 18%

Covenant spent two years proving that permissionless training could work at scale. The team’s Covenant-72B model ran across more than 70 independent contributors using only commodity hardware. Nvidia CEO Jensen Huang compared the project to Folding@home on the All-In Podcast. Anthropic’s co-founder cited the research. The work helped drive a roughly 90% rally in TAO during March.

Now that credibility is gone from Bittensor. So is the team, the models, and the research behind them.

What Covenant AI Alleged Against Bittensor

Dare’s post targeted Steeves directly. Steeves, known as Const, co-founded Bittensor and sits on its three-person multisig governance structure.

According to Dare, Steeves sold portions of his alpha holdings in Covenant’s three subnets: Templar (SN3), Basilica, and Grail. Those sales shifted emissions in a way that Covenant interpreted as economic pressure. The timing, amid operational friction between the teams, reinforced that view.

Dare also alleged that Steeves temporarily stripped moderation rights in Covenant’s Discord channels. The team said Steeves deprecated their infrastructure and community spaces without broader consensus from other stakeholders.

Dare called the multisig structure “decentralization theatre.” He argued Steeves retains effective control of the network while using the other signers as legal shields. “The power has never left one person’s hands,” the post stated.

Covenant concluded it could not keep raising capital or asking miners and investors to commit resources on a foundation one individual could undermine at will. The team said its research, models, and people would move elsewhere, with new project announcements coming soon.

Steeves Responds on X

Steeves addressed the claims within hours. He denied having the power to suspend emissions to specific subnets. According to him, he simply sold a small slice of his own holdings in subnets running near 100% burn code because they were not operational.

He also disputed the Discord claims. Steeves said Dare had pinned a note and posted about the changes himself. The moderation adjustment, according to Steeves, was temporary and aimed at stopping the deletion of honest criticism in Covenant’s channels.

Steeves said his sales amounted to less than 1% of what he had invested in those subnets. He framed the Covenant AI Bittensor exit as proof the network needed structural reform, not evidence of centralized control.

His proposed fix: lock-based subnet ownership. Under this model, subnet control would tie to time-locked tokens. Investors would gain visibility into team commitment. If an owner exits, the community could redirect conviction elsewhere.

Community Backlash Hits Both Sides

The Bittensor community reacted sharply. Many called the move a rug pull. Replies on X accused Covenant of cashing out after months of hype, then blaming Steeves to justify the departure.

One widely shared clip from a recent interview showed Dare praising Bittensor and the support it had given the project. Critics used it as evidence the exit was financially motivated, not principled.

Holders of subnet alpha tokens, particularly Templar, watched values collapse. Liquidity dried up as sentiment turned. Posts labeled the team “cowards” and “scam artists.” Some accused them of dumping on the miners and validators who supplied compute for Covenant-72B.

Supporters of Covenant countered that no team should stay in a network where one person can apply economic pressure without accountability. The debate remains unresolved.

TAO Price Takes an 18% Hit

TAO dropped from around $338 to as low as $263 within hours of the announcement. That 18% decline erased much of the March rally that Covenant’s work had helped fuel.

Trading volume spiked more than 150%. Derivatives traders liquidated positions while spot holders sold into the drop. The Templar token, tied to Covenant’s flagship subnet, also traded sharply lower in the same window.

On-chain observers and market reports attributed the selloff to two factors. First, the loss of a high-profile builder. Second, fresh doubts about whether Bittensor’s governance could deliver on its decentralized branding.

Loading chart...

What Comes Next for Both Sides

Covenant has not disappeared. The 72B model and the underlying research survive. The team says permissionless training was never exclusive to Bittensor. They plan to advance the technology independently.

For Bittensor, the next weeks will test whether the network can attract new builders to replace the lost emissions contribution. The lock-based ownership proposal from Steeves could address the IP portability risk, but it has not shipped yet.

The price action already priced in the builder loss. The community reaction is still processing the governance questions. Both sides have put their version of events on record. The market will decide which narrative holds.

This is not financial advice. Conduct your own research before making investment decisions.