Skip to main content

Bittensor's Two-Front Governance Crisis: Latent 11 Inherits the Codebase as TAO Bleeds $900M

· 11 min read
Dora Noda
Dora Noda
Software Engineer

In the same three weeks that Bittensor co-founder Const proposed rewriting the network's voting rights and Covenant AI walked away from its three flagship subnets, a quieter event reshaped the protocol's future even more profoundly: on April 2, 2026, the Opentensor Foundation transferred ownership of nine core GitHub repositories — including the Bittensor Python SDK and the btcli command-line tool — to a new entity called Latent 11.

The handoff was framed as decentralization. In practice, it concentrates control of Bittensor's only client implementation in a single new organization, at the exact moment the network's governance is unraveling. It is the rare crypto story where every plausible reading — bullish, bearish, and existential — depends on what happens in the next six months.

A Two-Front Crisis in 14 Days​

Bittensor's April 2026 timeline reads like a stress test:

  • April 2 — Opentensor transfers nine core repositories to Latent 11, ending its formal role as the foundation that originally created TAO.
  • April 9–10 — Covenant AI founder Sam Dare announces the team is leaving Bittensor, calling its governance "decentralization theatre," shuts down subnets Templar (SN3), Basilica (SN39), and Grail (SN81), and sells roughly 37,000 TAO (~$10.2M).
  • April 10 — TAO drops 25–27% intraday, from $337–340 to lows near $253–285. Approximately $650–900M in market capitalization evaporates and $9–11M in long positions get liquidated.
  • April 16–17 — Co-founder Const introduces the Conviction Mechanism (BIT-0011) at a town hall: a Locked Stake regime that ties subnet ownership and voting rights to time-locked Alpha tokens.

Both fronts — repository control and voting rights — are being rewritten simultaneously. That is unusual. Most chains weather one governance shock at a time. Bittensor is asking its community to absorb two structural changes while TAO trades around $242 with a market cap near $2.33B.

The Latent 11 Handoff: Decentralization or Concentration?​

The Opentensor Foundation has been gradually stepping back since February 2026, when Jacob Steeves moved to a standard participant role and the CEO position was eliminated. Governance shifted on-chain — validator consensus, subnet-owner participation, stake-weighted decisions — with no central entity holding override power.

Transferring nine repositories to Latent 11 fits that narrative. The foundation is no longer the formal upstream. The codebase has a new home.

The complication is that Latent 11 is now the only home. There is no Latent 11 plus three other independent client teams. The "Bittensor SDK" repo at latent-to/bittensor is the canonical source. btcli, the only mainstream command-line tool, lives at latent-to/btcli. Subnet operators, validators, and miners depend on this one stack.

Compare that to how other major networks handle core development:

NetworkCore development model
Bitcoin~6 lead maintainers across multiple jurisdictions, no single corporate sponsor, "minimal privilege" merge access
Ethereum5+ independent client teams (Geth, Nethermind, Besu, Erigon, Reth) coordinating via EIPs
SolanaAnza (Agave), Firedancer, and Jito Labs as competing client implementations under SIMD governance
SuiMysten Labs maintains the canonical stack — explicitly centralized for shipping speed
Bittensor (post April 2)Latent 11 owns the SDK and CLI; no second client team

Bittensor's model now sits closer to Sui's than to Ethereum's. That is a defensible architectural choice — fast, coherent, easier to ship — but it cuts against the "decentralized AI" narrative that drove TAO's rally. A handoff that looks like the foundation letting go is, in client-implementation terms, a handoff that concentrates upstream maintenance in a single new entity.

Why the Covenant AI Exit Hits So Hard​

Covenant AI was not a peripheral subnet. The team operated three of Bittensor's most prominent subnets — Templar, Basilica, and Grail — and built the 72-billion-parameter Covenant 72B model that had been the network's flagship technical achievement.

Sam Dare's exit accusations were specific:

  • Steeves allegedly suspended emissions to Covenant AI's subnets.
  • Steeves allegedly overrode the team's moderation authority over its own community channels.
  • Steeves allegedly unilaterally deprecated subnet infrastructure Covenant depended on.
  • Coordinated, timed token sales applied economic pressure during the dispute.

Whether or not every claim survives scrutiny, the framing — "decentralization theatre" from the team behind the network's most cited model — is the kind of thing investors quote in memos. The 25–27% drawdown in six hours was not just price action. It was the market repricing the credibility of the decentralization premise itself.

The harder question is what the exit revealed about Bittensor's de facto power structure. If a co-founder could squeeze even the network's most prominent subnet operator, the on-chain governance unveiled in February was thinner than advertised. That gap is what BIT-0011 is trying to close.

BIT-0011: Locked Stake, Decaying Conviction​

Const's Conviction Mechanism is a serious attempt at structural reform. The mechanics:

  • Anyone can challenge for ownership of a subnet by locking Alpha tokens for an extended period — months to years.
  • Each lock generates a conviction score that starts at the full locked amount and decays linearly to zero by expiry.
  • Every 30 days, the staker with the highest conviction EMA becomes the subnet owner.
  • Tokens stay locked until the interval concludes, eliminating cheap exits.
  • Initial deployment targets the three mature subnets at the center of the Covenant fight: SN3, SN39, SN81.

Two things are notable. First, BIT-0011 explicitly tries to make subnet ownership a function of time-weighted commitment rather than spot stake or founder privilege. That should make it harder for any single party — co-founder included — to apply economic pressure without skin in the game.

Second, it is being introduced under crisis conditions. Locked-stake systems rely on participants believing the network they are locking into will still be valuable months from now. Asking the community to believe that two weeks after a 25% drawdown and a developer-team walkout is a bet that the underlying narrative — decentralized AI as a credible alternative to hyperscaler models — survives this quarter.

The Two-Front Question​

Bittensor is now testing two independent hypotheses simultaneously:

  1. Can a single client team — Latent 11 — maintain core development without recreating the central-authority problem the foundation was supposed to solve? Polkadot's experience with Parity is the cautionary tale: client dominance created technical debt the Web3 Foundation took years to unwind. The optimistic counterexample is Sui, where Mysten Labs' tight grip on the stack has accelerated shipping without (so far) producing a governance crisis.

  2. Can a locked-stake conviction mechanism fix the de facto centralization that Covenant AI's exit exposed? This is essentially the question every governance reform in crypto eventually faces: do incentive redesigns actually change behavior, or do they just create new optimization games for sophisticated actors?

Both hypotheses interact. If Latent 11 ends up making decisions that subnet operators dislike, BIT-0011's conviction system gives stakers leverage — but only over subnet ownership, not over the client codebase itself. If Const wants the Conviction Mechanism enforced in code, Latent 11 has to ship it. The voting rights and the merge rights are now in different hands.

What's Holding the Network Together​

For all the drama, the operational picture is more resilient than headlines suggest:

  • 128 active subnets as of April 2026, up 97% from 65 at the start of 2025.
  • Estimated $43M in Q1 2026 network revenue, broadly unaffected by the Covenant departure.
  • First TAO halving completed, cutting block rewards by 50% and tightening new supply.
  • Grayscale raised TAO weighting in its AI fund to 43.06%, the largest single-asset reallocation Grayscale has made.
  • Spot TAO ETF applications filed by Grayscale and Bitwise on April 2, with an SEC decision expected by August 2026.

That is not the profile of a network in terminal decline. It is the profile of a network where the institutional bet ("decentralized AI is a real category") is decoupling from the operational drama ("our governance is messy and we're rewriting it in public"). The question is how long that decoupling holds.

What This Means for Infrastructure Builders​

If you operate or rely on Bittensor infrastructure — RPC providers, subnet operators, validator teams, agent frameworks integrating TAO — three concrete things just changed:

  • Single-client risk is now real. Subnet operators who previously assumed multiple independent maintainers would catch breaking changes need to track Latent 11's roadmap directly, not the Opentensor blog.
  • Subnet ownership is about to require capital lock-up. If BIT-0011 ships, "running a subnet" stops being a code-and-marketing exercise and starts being a treasury commitment. Plan accordingly.
  • The decentralization premise is actively contested. Anyone selling Bittensor exposure to institutional buyers should expect the Covenant exit and the Latent 11 handoff to come up in due diligence for the rest of 2026.

For multi-chain teams watching from the outside, the more interesting takeaway is that decentralized AI networks may have a structural reason to look more like Sui (one strong maintainer) than like Ethereum (many independent ones). Coordinating client diversity across rapidly evolving AI primitives is genuinely harder than coordinating it across a comparatively stable EVM. Whether that's an acceptable trade-off depends on what you think "decentralized" actually has to mean.

The Verdict Is in the Next Cycle​

The clean version of Bittensor's April 2026 story is "growing pains for an ambitious network." The honest version is that two of the three pillars of credible decentralization — independent client maintenance and resilient on-chain governance — are being rebuilt from scratch under live fire, while the third (broad subnet participation) is the only thing keeping the lights on.

If Latent 11 ships cleanly, BIT-0011 passes, and a second client team eventually emerges, April 2026 will read as the painful but necessary maturation of the protocol. If Latent 11 becomes a single point of failure, BIT-0011 gets gamed by sophisticated stakers, and Covenant-style exits repeat with other flagship teams, the same period will read as the moment Bittensor's "decentralized AI" thesis collapsed under its own contradictions.

The market is pricing both outcomes simultaneously: $2.33B market cap, 43% Grayscale allocation, a pending ETF decision, and a 25% drawdown all in the same month. That is what a binary in mid-resolution looks like.

BlockEden.xyz operates production-grade RPC and indexing infrastructure across major chains and increasingly across AI-agent workloads. As single-client networks like Bittensor concentrate maintenance risk, having a multi-chain infrastructure partner that monitors upstream changes across ecosystems becomes a load-bearing dependency. Explore our API marketplace to build on rails designed for an environment where governance and tooling can shift in 14 days.

Sources​

Share on Twitter