The team, which previously operated under the name Templar on Bittensor's Subnet 3, published a statement calling the network's governance model "decentralisation theatre." Founder Sam Dare wrote that the team "cannot continue raising capital, recruiting talent, or soliciting resources on a network where decentralisation promises can be overturned by a single individual." As of publication, neither Steeves nor the Opentensor Foundation had responded publicly to the allegations.

---

What Covenant AI Says Happened

Covenant AI detailed a series of actions it attributes to Steeves, who is known on-chain as "Const," during what the team describes as a governance dispute. According to the team's statement, Steeves suspended TAO subnet reward distributions to Covenant AI's miners, revoked the team's community channel permissions, shut down subnet infrastructure without consent, and applied economic pressure by selling large quantities of TAO tokens during the conflict. The team said governance decisions on the network lack transparent processes or community consensus, with actual authority concentrated among a small number of individuals within what they characterised as a three-person multisig structure. That characterisation sharpens their "decentralisation theatre" allegation considerably.

The allegation has a documented precedent. In January 2026, Steeves personally deployed 20,000 TAO into subnets, a move described at the time as "active ecosystem curation" that had already raised questions about founder-level influence over subnet outcomes.

The allegations land with particular force given Steeves' formal status. In February 2026, he resigned as CEO of the Opentensor Foundation, announcing the move as a leadership transition. His co-founder Ala Shaabana simultaneously stepped down as COO on February 13, 2026, with both departures together forming the announced leadership change.

His own words at the time offered a telling qualifier: "Not much will change, I'm still going to be in the same calls and do the same things...just not with the same legal authority (and that's important)." Covenant AI's account of events, which postdates that resignation, suggests his day-to-day influence over the network remained substantial.

---

Token Impact and Market Context

TAO fell from a range of roughly $325 to $340 before the announcement to an intraday low of $283.59, with CoinGecko recording a 24-hour decline of approximately 11.4% at the time of data capture. The token was trading at $286.37 with $1.48 billion in 24-hour volume. TAO's market cap stands at roughly $2.75 billion against a circulating supply of 9.59 million tokens.

The sell-off reverses part of a sharp rally driven by high-profile endorsements. In March 2026, TAO climbed nearly 90%, moving from around $180 to above $332, after Nvidia CEO Jensen Huang and investor Chamath Palihapitiya both praised decentralised AI training on the All-In Podcast. Huang called the approach "a modern version of folding@home," drawing a direct comparison to the distributed computing project Folding@home.

Bittensor's broader ecosystem token market cap reached $1.47 billion during that period. Covenant AI's own subnet token had gained 444% in the preceding 30 days. The governance controversy now casts doubt over whether that momentum can be sustained.

---

The Technical Stakes

The timing matters because Covenant AI's recent work represented a genuine technical milestone. In March 2026, the team released Covenant-72B, a 72-billion-parameter large language model trained across more than 70 independent contributors using commodity GPUs and standard home internet connections. The model scored 67.1 on the MMLU benchmark, placing it close to Meta's Llama 2 70B in performance. A key innovation called SparseLoCo reduced communication overhead between nodes by a factor of 146, making it practical to coordinate training across distributed hardware without expensive data-centre interconnects.

The arXiv paper documenting the work described the run as the largest decentralised LLM pre-training run in history. That description originates in the paper itself and reflects the authors' own claim.

The achievement drew institutional attention toward Bittensor, including a pending Grayscale spot TAO ETF application and European institutional access through a staked TAO ETP. These institutional developments occurred concurrently with the technical milestones; the precise causal relationship between them is an editorial inference rather than a directly sourced claim.

---

What This Means Outside the United States

The practical significance of Covenant-72B extended well beyond headline benchmarks. For developers in South Asia, West Africa, and East Africa, the model demonstrated that frontier AI training does not require hyperscaler compute budgets. According to DePIN market research, decentralised compute networks can offer cost savings of 60 to 75% compared to services like AWS or Google Cloud, which charge roughly $4.50 to $5.50 per hour for H100 GPU access. Comparable capacity on networks like Akash can run $1.20 to $1.80 per hour. These figures are market estimates drawn from DePIN industry analysis and should be treated as approximations rather than independently audited data.

For independent researchers and startups in markets like India, Nigeria, and Kenya, governance reliability is not an abstract concern. Sudden reward suspensions or infrastructure shutdowns (the specific failure modes Covenant AI described) are more damaging to teams without capital reserves to absorb the disruption. At India's AI Impact Summit in February 2026, Prime Minister Modi positioned India as "a central node in the global AI ecosystem," citing ambitions for over 38,000 shared national GPU units and $200 billion in data-centre investment.

The governance failure carries regulatory implications for those same markets. India's evolving TDS and GST crypto framework, Nigeria's Securities and Exchange Commission licensing regime, and Kenya's Capital Markets Authority proposals are each actively developing rules for digital assets and decentralised networks. Governance breakdowns of the kind Covenant AI describes provide ammunition to regulators who favour centralised oversight over decentralised alternatives, raising the stakes for platform credibility across the region.

Developers evaluating decentralised alternatives to centralised national or corporate cloud infrastructure need credible governance to justify the bet.

Covenant AI said it will announce its next project soon. Competing decentralised compute networks, including Gensyn, Render Network, and Akash, may attract renewed interest from developer teams reassessing their platform options.

The Opentensor Foundation had not commented as of publication.

Verse Press sought comment from the Opentensor Foundation and from Jacob Steeves directly prior to publication and received no response from either party. TAO price data is sourced from CoinGecko and reflects figures captured on April 10, 2026.