Bittensor Daily News
Latest editionSN91 Cascade merges safeguards for miner-funded rentals
Late-found October 31 merged PRs 392 and 393 add rental caps, failed-boot handling, shutdown cleanup and progress/deadline guards. These address the risk of machines continuing to cost miners when useful training is not progressing. The PRs describe code that still needs rollout: worker safeguards require rebuilt images, and a measured-time gate is off by default. No deployment or reduction in actual bills was independently observed.
Around Bittensor
8 more storiesSN8 Vanta reviews a possible stale-price paper-trading issue
The November 1 public brief attributes a stale-price execution complaint and planned hedging/arbitrage controls to channel participants and the team. It says the team was reviewing the report and did not confirm arbitrage. A separate digest gives a different flagged-account amount, so no loss total is published here. The underlying data and Discord originals were not verified, and no deployed fix or effect on another subnet is established. This remains a market-integrity lead requiring follow-up.
SN22 Desearch describes resumable public network-stat streaming
In its November 1 announcement, Desearch says its public network statistics stream updates once a second and can replay missed snapshots from the preceding 20 minutes using Last-Event-ID. The attached diagram illustrates a dropped connection followed by replay and resume. That is useful for dashboard recovery; reconnect behavior was not independently tested, and these statistics are distinct from the separate AI-search streaming endpoint.
SN26 Perturb wallet incident remains unresolved
Carry-forward safety watch: October 31 public coverage and an accessible screenshot attribute compromised wallets, unspecified losses and a product-announcement setback to a Perturb team message. The screenshot does not independently establish the speaker’s role, exact original date, attack cause, affected addresses or recovery. No new primary mitigation was found on the official X account. This is the same developing incident, not a new hack or an independently reconstructed chain exploit.
SN66 FReady brief reports a chat-log bonus and revised event plans
The November 1 brief says an earlier-proof chat-log bonus is live through ZIP uploads in the submissions menu. It also says a Caltech Mathathon arrangement was cancelled after event-structure changes and that an Imperial College London hackathon is planned for November 22–26. The original messages and counterpart event confirmation remain unverified; do not treat a planned event as a completed partnership outcome.
SN92 REAX documents paused competitions and demand-linked inference policy
Dated follow-up to the October 29 announcement and October 30 policy: the current official FAQ labels inference’s starting allocation 40%, Engine 20% and Generalist 30%, all measured within the miner portion of emissions. It says Generalist paused at October 30 16:30 UTC and Engine paused after its Friday slot, with no next crowning scheduled. Unpaid rewards are described as withheld/burned emissions, not qualifying explicit transaction burns. Further demand-linked adjustments remain plans; neither a fresh eTAO change nor lower realized selling pressure is established by commentary.
SN116 Carbon opens Control Room beta practice challenges
Late-found October 30 announcement: Control Room beta invites practice on physics-AI challenges. The inspected mining documentation identifies testnet 567, unpaid participation and no carry-over of records to mainnet. It lists 32GB RAM and four CPUs, with GPU optional. The post’s next-day testnet-weights plan is not proof of execution, and mainnet activation was not verified. The full 81-second demonstration remains a media-review gap.
SN117 EveryFrame lists Pika H3 Flex in its video catalog
The October 31 announcement adds Pika H3 Flex to the public video selector. The inspected catalog offered 5–15-second generation, picture and sound, and displayed 29 credits at five seconds during review. This is a visible catalog entry, not a completed paid render, quality benchmark or customer count. It is distinct from the earlier eleven-tool Director announcement. The full 27-second demonstration and its audio were not reviewed.
SN120 Affine publishes an accepted training-turn artifact after evaluation changes
Late-found October 31 evidence: the official turn 354 artifact records a green decision, candidate_promoted=true, a sample size of 179 and z=3.88046 against z_min=1, with batch and file references. This is the team-operated evaluation record, not an independent replay of the claimed model gain. Today’s public brief additionally attributes faster epochs and planned training-script disclosure to the team; those performance figures are not reproduced here.
Quick briefs & watchlist
9 updatesSmaller developments and early plans. Verification labels distinguish confirmed information from developing claims.
SN1 Apex brief describes submission-score viewing in its CLI
The November 1 public brief attributes to the team a CLI option for viewing scores of a miner’s completed URNetwork submissions. That can help participants inspect finished evaluations. The underlying Discord message and CLI flow were not independently tested; suggestions for more difficult tasks are discussion, not an announced new competition.
SN24 ms.computer changes its owner hotkey
Finalized block 9409822 (2026-11-01T08:24:12+00:00) records a successful SubnetOwnerHotkeySet at extrinsic 6, event 188. The stored hotkey changed from 5ELpkVn84ubfA7cYtz8YxL3NKYBSvMEZNGzkibkmsDe6YVcL to 5ELAvsvXL9kU9CMzkFcRiqFjw3QGVgA7MrvEfcHsicv5WvWX. The owner coldkey and registration height 2538424 were unchanged immediately before and after. This is an operational hotkey change, not a verified subnet sale or owner-coldkey transfer.
SN33 ReadyAI clarifies where grading occurs
The dated public brief says the challenge API distributes tasks while validators perform validation and grading. It also attributes a planned reduction in conversation-task share to the move toward a Skill Store. This clarifies the stated architecture; no new Skill Store launch or deployed workload change is established.
SN34 Yugos clarifies hosted evaluation handling
The November 1 brief says provider errors are retried, failed questions score zero without stopping a run, and the submission quota is three ZIPs per hotkey per 24 hours. It also says oversized calls are admitted despite per-question cost caps. These are attributed operational clarifications from the brief, not an independently tested security or cost guarantee.
SN44 Score removes three mislabeled evaluation blocks
The dated brief attributes removal of three faulty ground-truth blocks to the team: Petrol Station 9402900, PPE 9400500 and Beverage 9402000. This matters to evaluation fairness. Moving Vehicle detection to Profile 2 is only being considered for early next week, rather than already applied. Original channel messages and resulting scores were not independently checked.
SN61 RedTeam denies shutdown and points to future client news
The dated brief attributes a denial of shutdown to the team and says a client-progress update and another buyback tranche are expected. These are team statements and plans; no new client revenue or buyback transaction is independently verified here.
SN81 Reliquary verification overhead prompts founder commentary
In an October 31 reply about untrusted rollout verification, Const describes a cheap initial commitment followed by a more expensive asynchronous spot check and penalties for cheating. This adds explanation to the already-covered distributed-RL report, not a measured overhead number or independently demonstrated security guarantee.
SN107 Minos keeps IM2 launch and allocation plans provisional
The November 1 brief says the team has not finalized the IM1/IM2 allocation, with roughly 10% discussed as an initial share that could later change. It describes IM2 mainnet as forthcoming and intended to operate autonomously with an agent. Neither a mainnet launch nor an executed allocation change is verified.
SN114 SOMA discusses its next benchmark configuration
The dated brief describes medium effort, no Agent SDK system prompt, and compressor handling for subagent and side calls in the planned Claude Code and Haiku benchmark round. Cache-write accounting is still described as likely, and the backend was not expected in SOMA-benchmark before November 3. These are proposed settings and timing, not measured new performance results.
Stars indicate editorial priority, not investment quality. Routine competition updates receive at most three stars. Shizzy videos receive three; substantive non-video Shizzy coverage receives at least four. Scheduled events remain plans.
Sources & coverage notes
Audit trail
Coverage notes
Desearch describes reconnectable network data, Affine publishes accepted training-turn artifacts, and Cascade merges miner rental safeguards. Also: REAX’s documented pauses, EveryFrame’s model catalog, Carbon’s testnet beta and developing safety reports.
Coverage: Oct 31, 2026, 9:57 AM – Nov 1, 2026, 10:07 AM ET
Searched
- 128 current project profiles and dated public briefs; 138 repository paths and indexed public team posts
- Direct signed-in owner bookmarks/posts/reposts/replies, Const and Tensia sources; dated monitor receipts and unresolved queue
- Complete finalized allocation/burn and lifecycle review, all128 owner holdings, both public YouTube Videos/Live tabs
Public-data gaps
- Grok daily discovery, top-five attention and final omission requests hit the signed-in usage limit. Earlier dated monitor discovery/omissions are retained, but no current mention ranking is claimed.
- All 128 project profiles and dated brief URLs and 138 repository paths were collected; many briefs contain page shells. Indexed public X discovery and selected direct originals do not establish exhaustive access to every team timeline or private Discord.
- Discord-original access remains unavailable for several attributed reports. Vanta paper-trading pricing, Perturb wallet compromise and other unresolved safety claims remain developing; cause, losses and mitigations are not independently reconstructed.
- Selected media and primary artifacts were reviewed as recorded in the audit. Carbon and EveryFrame full demonstration videos and the long Yanez business video remain incompletely reviewed; no paid product tests, customer-revenue audit or independent benchmark replay was performed.
- Authenticated TaoStats cross-check and credit usage are blocked because no configured local key is available. Shared cached market data and finalized archive collection remain separate evidence sources.
Browse all 85 editions
Conference news · September 30, 2026
Exploit Conference Catch Up · Day 1
The teams, the ideas and the practical takeaways from Exploit Summit’s first day in Montreal. From Gamma and phone-based AI to robotics, science and the business of running a subnet.
Conference news · October 1, 2026
Exploit Conference Catch Up · Day 2
Catch up on Exploit Summit’s second day: Carbon’s Pitchtensor win, SN107 Minos’ research app, agent controls, robotics data, inference partnerships and the decisions behind building a subnet business.