Covenant AI's Withdrawal from Bittensor: A Wake-Up Call Revealing the Industry's Fake Decentralization

Covenant AI (formerly Templar) has spent more than two years building what is considered the most significant technological milestone in decentralized artificial intelligence to date: Covenant-72B — a language model with 72 billion parameters. This model was trained without licensing by more than 70 independent contributors on versatile hardware. This is not just a test run, but a major breakthrough that can be brought into production. This achievement was publicly recognized by Nvidia’s CEO on the podcast “All-In,” and it was also mentioned by Anthropic’s co-founder. Before that, it drove up the value of Bittensor’s TAO token by 90%, with some subnetworks reaching valuations close to $1.5 billion. However, right after that period of prosperity, the infrastructure underpinning everything became a weapon to attack them. In a public statement by Covenant AI, Bittensor’s founder, Jacob Steeves (pen name Const), unilaterally took several measures:

• Pausing the distribution of token rewards to the Covenant subnetwork;
• Reclaiming their governance rights over the community channel;
• Removing the subnetwork infrastructure without consulting anyone;
• Using token sales as a pressure tool in operational conflicts. Covenant AI pointed out succinctly that this is “centralized control disguised as decentralization.” Just a few hours after the announcement was issued, the TAO token plunged by more than 15%. As of the time of writing, its value is still down by about 9% over the course of the day. In response, Steeves stated that Bittensor would soon launch a truly independent subnetwork. However, this reply only affirmed—rather than refuted—the core issue: although Covenant is fully committed to developing it, the “independence” they need is still not guaranteed in the current architecture. Structural Trap: The Value Hostage Dilemma To understand why this is not just a one-off case, we have to look at the unique economic structure behind it—precisely the structure that makes the crisis both dangerous and predictable. Decentralized AI networks face a fundamental “cold start problem”: building real infrastructure—training tasks, model weights, contributor networks, and community trust—requires long-term investment lasting months or even years. This investment is paid for with time, capital, and reputation. But the final tokens that hold this value belong to the entire network, not just the builders. This creates a dynamic mechanism that we call the “value hostage”: The better you build it, the higher your network token value; the more valuable the tokens are, the stronger the influence the token holders have over you. And when you achieve your greatest success, that’s exactly when you’re most vulnerable. Creating value on someone else’s network means the value you generate can ultimately become a weapon against you. The more successful you are, the more you lose. This is not merely a governance failure specific to Bittensor, but a structural consequence common to all systems: when a minority holds veto power over critical authorities—such as token issuance, content moderation, and infrastructure upgrades—while claiming the system is “permissionless,” the seeds of the problems are already planted. The promise of “decentralization” is the operating foundation of the system—builders, operators, validators, and investors all make decisions based on it. Once that premise is proven false, or only “conditionally true,” economic losses won’t be limited to the parties involved, but will spread to everyone who trusted that story. Covenant AI points out that Bittensor’s governance structure is nominally a multi-signature mechanism with “co-management by three people,” but in reality, it is dominated by one person, while the other two are more like legal firewalls than actual decision-makers participating in governance. We cannot verify every accusation on our own. However, we can see the structural logic behind it:
• Multi-signature signed by a single primary owner is no different from a single private key;
• Governance processes can be bypassed unilaterally, which cannot be called governance;
• A token issuance mechanism that one person can pause is essentially a form of subsidy, not a protocol guarantee. This Is Not Just a Bittensor Issue, It’s a Widespread Industry Problem Some people may interpret the Covenant AI incident as a “Bittensor warning case.” This view is too narrow and even misleading. The deeper problem is that the entire decentralized artificial intelligence industry has long maintained a fragile balance between two opposing goals—rapid innovation and true decentralization. Fast iteration requires someone to make the final decisions; true decentralization means no one can unilaterally decide everything. Most projects default to the second option: they publicly claim decentralization while still retaining control, hoping that conflicts will never be exposed. And now, the truth has been exposed. In today’s decentralized AI ecosystem, a familiar pattern keeps repeating: A decentralized structure includes token distribution, community forums, and governance proposals; Wrapped within a centralized core—the founding team or organization that firmly controls the most critical parameters:
• Token issuance schedule
• Protocol upgrade rights
• Subnetwork acceptance mechanism
• Community management rights This does not necessarily stem from malice. Early networks require close coordination, and fully on-chain governance for technically complex AI infrastructure has not yet been solved. However, the gap between public promises and actual power creates a kind of “structural debt.” Covenant AI’s experience is exactly what this debt looks like when it comes due. In AI networks, this mechanism is more dangerous than DeFi or Layer 1 because the builders’ investment stakes are higher than ever. Training a model with 72 billion parameters is not a two-week sprint project, but a long-term, extremely costly, time-consuming, and reputation-risk battle. By the time Covenant AI realized the severity of the governance issues, they had already completed the most important part—and that’s why they became a target. This asymmetry is extremely brutal:
• The internet can operate at any time; However, builders cannot “cancel” a completed project. Similar events will repeat if the following three conditions are met:

1. Builders need to invest a significant amount of time and resources;
2. Tokens can capture the value of these investments;
3. A minority of people can still unilaterally interfere with governance. These three conditions are extremely common in current AI networks. Therefore, we should not expect fewer “next-generation Covenant AIs,” but instead anticipate more. True Decentralization Demands What? When faced with events like this, the industry usually falls into two extreme reactions: One approach is to praise any project that claims to be “decentralized” without any critical scrutiny; The other approach is to completely deny the feasibility of decentralization, treating it as a scam. Neither option is viable. We want to clarify one thing: achieving true decentralization in AI infrastructure from a technical standpoint is extremely difficult. Anyone who tells you it is “easy” most likely has never really tried.

• Cooperative training among regional nodes
• The ability to verify computational work
• Token-based incentive mechanisms against manipulation These are still unresolved issues or only partially addressed. An honest industry should acknowledge this. However, technical difficulty does not mean it’s impossible, nor should it be used as a pretext for misleading propaganda. We propose a simple yet sharp standard: Can the infrastructure you rely on be exploited to work against you? If the answer is “yes,” then no matter how the white paper describes it, and no matter whether there are governance votes, this decentralization is merely superficial. This question cuts through all the noise. It doesn’t care whether you have a DAO or a community forum. It only asks one real question: In the event of a conflict, can a small group of people unilaterally suspend your rewards, cut off your access, or force you financially? If so, then the claim of “decentralization,” no matter how sincerely the original intent was, lacks a structural foundation. The real solution is to integrate decentralizing constraints into the protocol layer, rather than relying on the “goodwill” of any individual or group. This means:
• The token issuance mechanism is defined by on-chain rules and cannot be unilaterally suspended;
• Upgrades require genuine consensus from all parties, not just formal approval;
• In governance, those who truly contribute— (builders, computational power providers, and validators)—have a voice proportional to their contributions and are protected accordingly. Building in this way is harder, slower, and seems less efficient in the short term. But only in this way can the promise of “building without permission” truly be credible. For example, using a Proof-of-Work (Proof-of-Work - PoW) model to allocate computational resources can establish governance weights directly linked to contribution: one unit of computational power is equivalent to one vote. It does not rely on ownership capital or the subjective judgment of the founding team. It anchors power to something that cannot be created or revoked centrally—namely, verifiable computational work. This cannot solve all governance issues, but it provides a starting point for resisting censorship and manipulation. Covenant AI: What Has It Proven? And Where Should Its Future Head? We should not let governance failures overshadow its technological achievements. Covenant-72B is a real breakthrough. Before Covenant AI, training a 72-billion-parameter model on more than 70 independent nodes, versatile hardware, and decentralized infrastructure was widely considered impossible. They have proven that decentralized training is fully technically feasible. Then, the internet betrayed them, but that cannot erase the truth. The real question is pushed to the next stage: Now that we know this technology is feasible, how do we design a governance structure that can keep this model sustainable in the long run? How can a talented team like Covenant AI, after many years of hard work, still be confident that the infrastructure they rely on will not turn into a weapon against them? Covenant AI has stated that it will continue its work after the Bittensor project. This is true in some sense. Training decentralized AI is not an exclusive feature of Bittensor; it is an independent technological capability applicable to any network. Subnetworks can disappear, but the technology itself cannot. But this industry cannot stop at just “attracting excellent teams that have escaped poor governance.” We must build a governance structure that eliminates the need for people to flee. Covenant AI’s project is an expensive “idea experiment.” It does not validate decentralized training (has been proven). The question is what will happen when governance design lags behind technological ambition. This lesson is very clear—everyone just needs to listen. The question is: Is the entire industry ready to face it?