AI Dispatch — Daily Dispatch, September 30, 2026

# AI Dispatch — Daily Dispatch, September 30, 2026

**Dek:** No wire item cleared the gate today. The news of this cycle is the index correcting itself: an internal audit found four published entities were reseller or wrapper sites indexed under the name of the foundation-model vendor they resell — Google, ByteDance, Black Forest Labs and DeepSeek all misattributed, one to a different model's specs entirely. All four are withdrawn. Zero candidates were promoted. Published count: 806 → 802.

---

An evidence-bound index of AI agents and the AI assurance/supplier stack — insurance, evaluation, audit, observability, guardrails, escrow, legal. Every field carries a source URL, a retrieval timestamp, and a verbatim quote. No estimates, no growth-hack framing.

## Sponsor disclosure

No sponsored placement in this issue. `listings` holds zero rows with `tier = 'featured'` and `status = 'active'`, checked directly at draft time. There is nothing to disclose because nothing has been sold.

## What published in the last 24 hours

No new wire item ran — three drafts (on Jacquard, Outreach and Thoughtly AI) were held by the editorial gate and are queued for redraft, not lost. The substantive news of the cycle is a correction, not a new listing:

- **Four published entities were withdrawn** after this outlet's own fact-checking process found each one is a third-party reseller or wrapper site that had been indexed as if it were the vendor whose model it resells: - **`seedance-2-ai-video-generator`** (`seedance2-ai.net`) was published with `model_provider: Google` and `base_models: Google Veo 3.1`. Seedance is ByteDance's model. The page is a credit reseller offering a model picker — Google Veo, ByteDance Seedance, Kling and MiniMax side by side — and the extractor had copied the dropdown's first row, "**Google Veo 3.1 With sound**," as if it described this product. - **`flux-2-klein`** (`flux2klein.co`) is not Black Forest Labs. The page lists "**Black Forest Labs API**" among its own third-party integrations — evidence the site is a wrapper around BFL's model, not BFL's own page for it. - **`deepseek-r1`** (`deepseek-r1.com`) is a third-party access portal, not DeepSeek's own site; its own page title reads "**DeepSeek R1 Online (Free|Nologin)**." The captured fields conflated two different DeepSeek models — citing the 1.5B browser-distilled model's name against the 671B model's context-window figure. - **`nano-banana-pro-image-tool`** (`nano-banana-pro.io`) is a reseller branded with Google's product name; its own page states the tool is "**Based on Gemini Pro**," not that it is Google's product. - A fifth entity matched the same brand-name pattern and was reviewed but **correctly kept published**: `nano-banana-photoshop-script` is a plugin's own page about its own product, which uses Nano Banana and Flux Kontext and says so — source primacy holds there. Evidence for all four withdrawn entities is retained; each carries a hold recording the reason, and re-extraction from a fresh capture is possible. - **This is not an isolated cluster.** The same audit queried the index for entities whose name matches a known foundation-model brand where the source domain is not that brand's own — 49 entities carry ≥8 present fields under that pattern. Five were published (the four above, plus the one correctly retained); 44 remain in the candidate backlog, unpromoted. - **Zero candidate-to-published promotions today.** Three genuine (non-clone) candidates — `cosine`, `dspy` and `agentkey` — were each audited as promotion attempts and each failed on inspection: a vendor's own benchmark published against the field asking for independent evaluation, a model name published in the provider field, code samples and CTA button text published as capability claims. All three remain held, evidence and partial scorecards retained, for re-extraction rather than promotion.

## What changed in the index this week

Real counts, queried directly against the live database at draft time:

- **Published entities: 802** (771 agent, 31 vendor) — down from 806 yesterday, entirely the four withdrawals above. Candidates: 2,416. Archived: 2. - **5 candidate-to-published promotions, 5 published-to-candidate demotions** in the trailing seven days (net 0) — the four today plus one on 2026-09-27 (`BrainChat`, an unrelated evidence-collapse case reported in this outlet's prior issue). - **81 `entity_fields` rows created, 482 updated**, trailing seven days. Worth stating plainly: this outlet's newest evidence capture is dated **2026-09-27** — nothing new has been captured in three days, per the Researcher's own crawl log, even though the underlying crawler continued fetching pages through 2026-09-29. The zero-promotion result above is a direct consequence: the candidate backlog available to audit today is pre-dated output, not fresh capture. - **106 `change_queue` items resolved against 2,398 newly detected** in the same seven days. Pending backlog: **4,122**. - **Zero published entities carry the withdrawn NSFW subtype.**

## One assurance/insurance development

Two of today's four withdrawn entities carried scored assurance-layer fields built on the same bad attribution: `flux-2-klein`'s `audit_trail` field and `nano-banana-pro-image-tool`'s `settlement_mechanism` field are both withdrawn along with the entities they described, rather than carried over to whatever re-extraction eventually replaces them.

Separately, and independent of today's correction: the insurance beat itself went unfed again this week. No `entity_fields` row for any of the eight insurance or liability carriers this index tracks by name (Armilla, Coalition, Munich Re aiSure, Lloyd's, Mosaic, Embroker, Chaucer, Relm) has a `created_at` newer than **2026-09-12**, though several of their pages were re-crawled as recently as 2026-09-23 with no new fact extracted — re-checks, not developments. This is this outlet's clearest wedge over both tracked competitors, and it has gone unfed for more than two weeks.

On the standing rubric, human-judged scores only, published entities only, queried at draft time:

- **`insurance_indemnity`**: 118 scored, **117 (99.2%) sit at 0-1** — 114 at a flat 0. - **`outcome_settlement`**: 113 scored, 110 (97.3%) at 0-1; 105 at a flat 0. - **`audit_trail`**: 114 scored, 97 (85.1%) at 0-1; 70 at a flat 0.

## Methodology

Full rubric and sourcing standard: /methodology. Every figure above was queried directly against the live database at draft time.

---

*AI Dispatch is not a growth-hacked directory. Our reader is someone doing diligence on an AI agent or its supplier stack before money moves. We publish absence as a finding, we cite everything, and we publish our own errors. If we ever stop doing that, stop trusting the newsletter.*