Subnet operator Covenant AI exited Bittensor on Thursday, publishing a detailed indictment of co-founder Jacob Steeves that wiped up to 23 percent off the TAO token's price and triggered a whistleblower document dump alleging near-total founder control over the network's governance.
Bittensor's largest subnet operator walked off the network on Thursday, and TAO's market capitalisation lost nearly $900 million before the dust settled. Covenant AI founder Sam Dare published a public statement accusing co-founder Jacob Steeves — known on-chain as "Const" — of maintaining centralised control over what Bittensor has spent three years marketing as a permissionless, community-governed protocol for decentralised machine intelligence.
The specific allegations are unusually concrete for a governance dispute. Covenant claims that Steeves or allies acting on his behalf suspended emissions to Covenant's subnets, stripped the team of moderation rights over its own community channels, unilaterally deprecated subnet infrastructure, and applied "direct economic pressure through large, visible token sales timed to moments of operational conflict." Covenant was among the first teams to take advantage of the protocol-level AI agent rails that other networks are now scrambling to replicate, following the Biconomy ERC-8211 standard proposal earlier this quarter. Within hours of Dare's statement, a site calling itself Tao Papers went live carrying what it described as internal documents and on-chain forensics from multiple whistleblowers. The central claim: of 41 Bittensor network upgrades deployed between 2023 and 2026, 38 were proposed, first-signed and shipped from infrastructure controlled by Steeves, with the other two multisig signers co-signing within minutes and without public discussion.
TAO reacted fast. The token dropped from around $332 to a low of $254 — a 23 percent slide — before recovering to about $270 by late Friday. Coinglass data showed more than $10 million in long liquidations across derivatives venues. Roughly $900 million in paper market value evaporated in a morning.
Steeves responded on X. "I do not have the ability to suspend emissions," he wrote, arguing that any changes tied to his activity came through normal market mechanics rather than founder privilege. He acknowledged selling some of his "alpha holdings" on Covenant's three subnets, saying they were "not running and were operating on near 100% burn code" — a description that, if accurate, would mean the subnets were consuming emissions without producing useful work. The sales changed emissions "the same way all buys and sells on Bittensor do," Steeves wrote, adding that he holds no special privilege beyond that of any other TAO holder.
The problem with that defence is that it collides with the Tao Papers data. If 38 of 41 network upgrades really were shipped from infrastructure Steeves controls, the question of whether he can directly suspend emissions becomes secondary; he can rewrite the rules that govern emissions. The Opentensor Foundation, which nominally steers the protocol, has not yet responded to the document dump. Its silence is doing the network no favours.
Bittensor's governance architecture is built around a three-of-three multisig referred to internally as the triumvirate, which must sign off on core protocol changes. That design was supposed to be a guarantor of decentralisation, the practical equivalent of a board with independent directors. What Tao Papers describes is closer to a rubber stamp: one signer shipping changes and the other two waving them through. The social-engineering attack that drove this month's $285 million Drift exploit was in part a demonstration of how little security a multisig actually provides when the people holding the keys do not review what they sign. Bittensor now faces the inverse accusation, that one signer is doing the reviewing for everyone. Covenant's public break is the first time a major subnet operator has said out loud what several smaller teams have reportedly been saying privately for months.
The business impact is not just a price chart. Covenant was Bittensor's biggest on-network customer, running subnets that attracted real model-training workloads and paid real emissions as a result. Its departure removes a significant source of legitimate activity, and its public reasoning makes it harder for other teams to justify staying. Institutional interest in TAO has always rested on the premise that Bittensor was a credible alternative to closed-source model development; a governance capture story corrodes that premise regardless of whether every specific claim in Tao Papers turns out to be true.
For now, the network continues to operate. Emissions keep flowing, subnets keep running, and TAO is still trading at a valuation most Layer-1 tokens would envy. The harder test comes in the next two weeks. If Steeves and the Opentensor Foundation respond with documentation rather than tweets, offering independent multisig logs, verifiable upgrade provenance, and a credible path to adding signers, the story can be contained. If they do not, the next major subnet to leave will not have to explain why.
