Dates and verification labels belong to the original edition. A past verified announcement can now be outdated. Historical project names are retained; an SN filter searches the slot across registrations. All keywords must match the story text or subnet label. Ratings describe editorial priority, not investment quality.
Bitsec reports finding and patching a critical vulnerability beyond smart contracts
Bitsec says its tooling found a critical remote-code-execution vulnerability in an open-source project, and that the issue was reported and patched. The September 8 post presents this as evidence that its smart-contract audit tooling can extend into general software security. The announcement is first-party, but its text does not identify a public advisory, affected version or independently checkable patch. The finding and remediation therefore remain Bitsec-reported.
A reported critical vulnerability discovery is a meaningful security and product-expansion story. The absence of a linked advisory limits independent confirmation.
SN119 changes from Satori to CapabilityForge Next
The current subnet identity names SN119 CapabilityForge Next and points to a new repository published September 9. Its initial design rewards training-data improvements for learning-oriented AI, starting with mathematics and natural sciences. Miners would submit traceable data patches; validators would test whether they improve a frozen baseline, including accuracy and age-appropriate explanations. The initial repository contains the project description and images, not an implemented mining network. The identity and proposal are confirmed; live operation is not.
A new name and research mission replace the prior Satori identity. The early documentation stage is essential context for a rebrand this substantial.
UMI takes Deaf-accessibility discussions to a congressional office
UMI says it met with a congressional office on September 8 to discuss Deaf accessibility, healthcare and emerging AI. The team says the conversation opened introductions to disability-rights, healthcare and national accessibility groups. This is a reported outreach meeting, not congressional endorsement, a government contract or a policy change. Separately, its published sign-language demonstration shows an imperfect early translation with a 71% semantic-similarity score for one example, rather than a general accuracy benchmark.
Policy and distribution outreach can matter for assistive technology. The story is prominent while the meeting’s actual scope and the limited demonstration remain clear.
UMI prepares seven-day bootstrap rewards, with chain cutover still gated
New UMI source implements a temporary seven-day service-weight mechanism that gives eligible public-endpoint miners equal weight. It explicitly does not rank translation quality or activate translation rewards. September 9 transition documentation says the first submission remains blocked until legacy weight rows are inactive and pending commits have cleared on finalized chain state. UMI also scheduled a development AMA for September 9 at 17:00 UTC / 1 PM ET.
The bootstrap and transition conditions directly affect miner participation. Reporting a fully activated translation economy would overstate what the source establishes.
SOMA says its new compression algorithm cuts token costs by up to 15%
SOMA announced an algorithm improvement of five percentage points, taking its advertised savings to as much as 15%. The September 8 post links directly to the application. The result is the project’s published claim and will depend on the workload; it is not a promise that every request or complete AI bill falls by 15%. This is a new improvement to the context-compression product, rather than a repeat of its earlier launch.
A concrete improvement to a usable cost-saving product is commercially relevant and easy for users to evaluate.
Cortex opens a new research site — Proof intake ready, automatic payouts still pending
Cortex’s new network.cortex.foundation site describes two Proof research topics and says submissions are being accepted. The same page states that judging is partial, submissions are held in memory, weights are not yet sealed and automatic payouts are not wired. Bounty scoring is paused, with reports refused until its scoring feed connects. New repository work adds isolated research execution and rejects empty research artifacts. The site is available, but its own status does not support a fully operating paid research marketplace.
The new research direction and public intake are significant. Clear status distinctions prevent the prominent “live challenges” wording from being mistaken for active payouts.
Minos’ emission share expands to roughly 7.5% after restoration
Yesterday’s report confirmed that Minos was receiving emissions again. The new 11:13 AM ET snapshot shows its share at 7.522%, up from 1.181% in yesterday’s saved snapshot. That is a further allocation change, distinct from the initial restoration. The latest channel brief still describes a miner-burn reduction as coming soon; this scan did not establish that the burn has already been removed.
A large increase in network allocation after the pause is a material follow-up. The still-unconfirmed miner-burn step is kept separate.
OpenRoboto fixes stale fee documentation that could cost miners a submission
OpenRoboto’s September 9 documentation change removes hard-coded entry fees and directs miners to the current competition API value. The commit explains that the simulation fee moved from 0.1 to 0.2 TAO on September 3 while some guides stayed stale; it also corrects real-hardware guides that quoted 2 TAO instead of 1.5. Sending the wrong amount could burn funds without an evaluation. These are the values described in the fix, not a permanent fee schedule: check the live competition before submitting.
A documented fee mismatch can directly cost miners money and prevent evaluation, making this more important than a cosmetic documentation edit.
OpenRoboto discusses European robotics deployment with István Joó
OpenRoboto reports a meeting with István Joó covering realistic early robot jobs, specialized manipulation policies and work with European universities and factories. The team says more European updates will follow. This is a first-party account of a business-development conversation; no signed customer, factory deployment or university partnership is announced in the post.
New commercial outreach is worth tracking, while the meeting is kept distinct from a completed deal.
ReadyAI brief points to an August buyback 2.25 times the prior month
The latest ReadyAI channel brief reports that the August buyback is planned for this week, at 2.25 times the prior month’s amount, using 75% of dashboard revenue. It also describes a monthly cadence and an AMA next Tuesday, September 15. A direct team announcement and completed buyback transaction were not located during this scan, so this is a developing scheduled event, not a completed purchase.
A larger revenue-funded buyback could be material, but the amount and execution still need primary confirmation.
MVTRX brief reports v0.6.0 mainnet and a noncustodial swap tool
The September 9 MVTRX channel brief reports a v0.6.0 release covering agent framework and scoring changes, plus a noncustodial swap tool. It describes v0.6.1 scoring work as the next testnet step and an AMA for today. The product and release claims were not independently matched to a dated primary release during this scan; they remain developing rather than a verified launch.
A mainnet release and user-facing swap product would materially expand the project, warranting a prominent developing item.
Capcomp remains sharply lower even as its emission share turns positive
Capcomp’s rolling 24-hour α/TAO change is -39.16% at 11:13 AM ET, following the sharp decline covered yesterday. Its observed emission share is now 0.138%, compared with zero yesterday. The positive allocation does not by itself establish normal miner payouts. Uncorroborated community allegations are not being used to explain the price movement.
Continued severe price weakness alongside an emission return is a material market follow-up. No unsupported cause is assigned.
Apex updates Humanoid Box Scramble timing, physics and partial-progress scoring
Apex’s Humanoid Box Scramble competition configuration now specifies one-day rounds and a one-day model-reveal period. Other merged changes make room and crate friction authoritative in robot contacts and score timed-out runs on partial progress. The pinned competition image advances to v0.1.5. These are published competition-rule and implementation changes; the source alone does not confirm every validator is running the new image.
Round timing and scoring affect miner strategy, while the physics correction changes what the robot is actually evaluated against.
DSperse v14.14.3 guards against accidental full-burn weight commits
DSperse’s new guard holds the previous weights when all non-owner miner scores are zero. The old check missed this situation when the owner still had a nonzero weight, allowing that entry to normalize to the full allocation. The fix is merged into the v14.14.3 release. SN2’s observed subnet emission share is zero at 11:13 AM ET; the code change is not evidence that network emissions or miner payments have recovered.
A reward-setting safeguard addresses a specific failure mode, while current emission status prevents a premature recovery claim.
Teutonic expands evaluation from 4,000 to 20,000 samples across five datasets
Teutonic’s merged configuration replaces its prior three-dataset mixture with five sources spanning general text, pretraining behavior, math, code and supervised reasoning. The evaluation sample count rises from 4,000 to 20,000. Today’s team-channel brief says the new mix still needs pretokenization before live evaluation, so the merged configuration should not be read as completed activation.
A broader and larger evaluation set changes the standard miners must optimize against.
Allways publishes a SOL-to-native-TAO walkthrough for BUTT and CONST users
Ventura Labs has published a walkthrough showing how Solana users can reach native TAO through Allways. Its September 8 post says a swap uses one credit and new accounts receive two free credits. The new guide specifically addresses the BUTT and CONST audience, creating a clearer route from Solana-based meme activity to Bittensor. No swap was performed as part of this report.
A concrete onboarding guide connects a newly active audience with native Bittensor access.
Vanta describes bank-transfer and USDC settlement on one public ledger
Vanta says eligible rewards can settle by bank transfer with a Stripe reference or by an on-chain USDC transfer with a public transaction hash. Both routes are recorded on its public ledger. This clarifies settlement options; it is not evidence of a new Stripe partnership or a guarantee that every participant is eligible.
Useful product-operation detail, with partnership and eligibility claims kept within the announcement.
Blockmachine reports stronger September signup momentum after capacity upgrades
Blockmachine says September signups are on pace to double August’s total, nine days into the month, following capacity and performance upgrades. This is a company-reported run rate, not a completed monthly result; no absolute signup count or revenue figure is disclosed. The announcement is a useful demand signal for the inference business, with the measurement limit explicit.
Reported customer acquisition and capacity improvements are commercially relevant, even before a complete monthly result is available.
AdTAO makes its reward-curve tail a dated, published parameter
AdTAO’s new source makes the tail of its reward curve an explicit switchable parameter with an effective date. A companion tool previews how a day’s weight vector would change under another tail using a copy of the ledger. This is an operational transparency and reward-design update, separate from the daily-payment and training-data changes already reported yesterday.
The update makes proposed reward changes easier to inspect before they affect contributors.
UR records a 544,553 α conviction lock and a separate 50,000 α lock reduction
SubnetRadar’s September 9 chain-event records show 544,553.077 α added to a lock targeting UR’s owner hotkey at block 9,027,997. A separate hotkey’s locked amount then fell by 50,000 α at block 9,028,349. These are two distinct events, published at 06:03 and 07:13 UTC. The reduction record identifies a change to the lock, but does not establish that those tokens were sold or specify whether a transfer, forced reduction or closure caused it.
The scale of the commitment and the separate reduction make this a material token-lock update. The records are reported without inferring a sale, wallet identity or motive.
Searches published headlines and summaries, not private bookmarks, unpublished leads or source-page contents. Duplicate headlines within an edition appear once; separate editions remain separate so follow-ups are preserved. Share the page URL to retain your filters.
