$TAO dropped from $337 to $270 in under 24 hours on April 10, 2026, after Covenant AI announced it was leaving the Bittensor network. The exit came with a 1,200-word statement from Sam Dare, Covenant AI’s founder, describing the network’s governance as something built to look decentralized while operating as anything but.
Dare’s accusation was direct. He said Bittensor’s claimed decentralization is not just incomplete. It is a lie.
One Man, One Network
The statement published by Covenant AI on X names Jacob Steeves, Bittensor’s co-founder known online as Const, as the person who made the decision to choke the team’s operations. According to Dare, Steeves suspended emissions to Covenant’s subnets without warning, effectively cutting off their revenue stream.
He also alleged Steeves stripped the team’s moderation rights inside their own community channels, deprecated subnet infrastructure without any formal process, and executed large, visible token sales during the conflict as direct financial pressure.
“When a single actor can suspend a subnet’s emissions, override an owner’s authority over their own community spaces, publicly deprecate projects without process, and use token sales as a coercive mechanism to compel compliance, that is not decentralization,” Dare wrote. “It is centralized control with decentralized branding.”
Bittensor’s governance structure, Dare explained, uses a three-person multisig model that the network presents to the public as distributed control. He said it is not. He described it as a structure where Steeves retains effective authority and the other two individuals function, as Dare put it, as legal shields who can be held accountable while Steeves stays insulated.
What Covenant Built, and Why That Matters
Covenant AI was not a minor contributor. The team trained Covenant-72B, a 72 billion parameter model across more than 70 independent contributors running commodity hardware. That run, completed permissionlessly, became the largest decentralized LLM pre-training run on record.
The project attracted attention from some of the most influential voices in AI. Nvidia CEO Jensen Huang publicly praised the work. Anthropic’s co-founder acknowledged it. That institutional recognition sent $TAO on a 90% rally in March 2026. As Coin Bureau noted on X, the same centralization claims that triggered the exit pushed $TAO from $337 to $270 within a single day.
The irony is structural. The praise that validated the project’s decentralized credentials is now making the departure more credible, not less.
Steeves Pushes Back
Jacob Steeves responded directly on X, disputing each point Dare raised.
On the emissions allegation, he wrote: “I do not have the ability to suspend emissions. What I did do is sell some of my alpha holdings on his three subnets, because they were not running, and were on near 100% burn code.”
Steeves argued the change in emissions came from his activity as a regular token holder, not through any administrative privilege. He said he has no powers beyond what any ordinary TAO holder has. On the moderation claim, he said Dare had been deleting legitimate community criticism inside his own channel, and that he temporarily blocked that action before reinstating it. He did not remove Dare’s moderator role, he said. He simply stopped him from suppressing other users’ posts.
He also dismissed the token sales as small, less than 1% of what he had invested in Covenant’s teams, and said visibility was unavoidable given his position.
$9.1 Million Liquidated, $880 Million Gone
The financial damage was fast. According to data cited by Bitcoin.com News, the crash triggered $9.1 million in long liquidations on April 10. Trading volume surged to $1.72 billion, compared to the $500 million average seen earlier in the month. Total market cap loss across the Bittensor ecosystem came in at approximately $880 million within 12 hours.
Subnet tokens tied to Bittensor’s staking mechanics saw even steeper falls. Some dropped over 55% on the same day. Covenant AI’s founder also sold his entire subnet alpha holdings, a wallet that held more than 37,000 TAO tokens, in the process of departing.
A Structural Question Without an Answer
The conflict puts something bigger than one team’s exit on the table. A newly launched site called Tao Papers published on-chain data the same day, claiming that of 41 network upgrades executed between 2023 and 2026, 38 were proposed, signed first, and deployed from infrastructure controlled by Steeves. The remaining two multisig participants reportedly co-signed each upgrade within minutes and with no public discussion recorded.
Dare’s position is clear. Covenant AI built something the world said was not possible and proved distributed AI training works at scale. He said the research, the models, and the team are leaving with him, and that announcements about what comes next are coming soon.
Steeves, for his part, pointed toward “headless” subnets as the next structural step for Bittensor. A design where subnets run autonomously, without needing anyone’s permission.
Whether that answer satisfies the question Dare raised is something the network will have to reckon with.












