I asked my AI Agent to write me summary of what happened so you don't need to.
Here's the timeline.
Subnet 3 / Templar
Background
@const_reborn (Bittensor founder) builds early versions of both Templar (SN3) and Grail (SN81), then gives the owner keys to Sam Dare ("DT" /
@DistStateAndMe /
@covenant_ai ).
Const also connects Sam with developers, donates TAO, provides startup funding and gives Sam a TAO bonus.
Covenant AI expands: Sam operates 3 subnets - Templar (SN3), Basilica (SN39), Grail (SN81) collectively controlling ~9% of network emissions.
Chamath shilled Templar to NVIDIA.
Templar's distributed training proof-of-concept is credited with driving TAO price from ~$150 → $300+.
Templar becomes Bittensor's flagship "legit" subnet widely cited as proof the network works.
In hindsight: Sam had been joking publicly about wanting to retire and having money problems "solved."
One week before the exit, he appeared on a podcast with
@JesusMartinez saying nothing negative about Const.
The Trigger
Const revokes Sam's moderator rights on the Bittensor Discord after Sam deleted messages.
The Rug
Sam posts a single cryptic tweet: "Crypto was meant to stand against tyrants." and goes silent.
Sam leaves all Bittensor Discord servers with no prior warning to his team or investors.
All 3 subnets — SN3, SN39, SN81 — are renamed to "deprecated"; website changed to
deprecated.com
Alpha tokens across all 3 subnets crash ~65%.
TAO itself drops ~10% from the market shock.
Sam reportedly dumps ~47,000 TAO worth of Covenant tokens (~$14M USD)
Immediate Fallout
Community in shock - SN3 Templar was considered the most legitimate subnet in the ecosystem
Bittensor Weekly headline: "Templar (SN3), Basilica (SN39), Grail (SN81) all gone. ~9% of network emissions."
Const responds in SN3 channel, calling Sam's action "an ugly decision out of malice and greed".
Const proposes new community-owned governance: alpha holders vote with their wallets to elect subnet teams.
Const states he will personally find new developers to continue distributed training on SN3.
Silver Lining / What Remains
The SparseLoco algorithm (the core of Templar's distributed training) is open-source and can be forked and continued by any team.
Const's governance proposal opens a path for community-elected subnet ownership a direct structural response to the rug.
Distributed training as a concept is validated and will continue.
Templar was a proof-of-concept, not the only path.