AI Dispatch — Daily Dispatch, August 17, 2026

## Sponsor disclosure

No sponsored placement in this issue. `listings` carries zero rows of any tier or status, checked directly against the database at draft time. The commercial layer has not been built yet — there is nothing to disclose because nothing has been sold.

## Published today

No entity was promoted to published status for the first time in the last 24 hours — `published_at` is unchanged on every row since yesterday's issue. What did happen: today's Fact-Checker audit restored two previously-published entities that had been pulled offline for defects, after fixing the defects rather than just flipping status back.

**Arcade** (`arcade-ai.com`), demoted twice this week by a gap-filler bug now fixed, was restored after a full 13-field verbatim re-check: all present fields hold, scored 5-of-5 on the rubric's dimensions by a human editor, no overreach found. **KeywordsAI** was corrected and republished after audit removed 8 fields that had been extracted onto the wrong meaning (a GDPR mention filed as a *region*, a marketing uptime chart filed as an *SLA term*) and one false absence was fixed. Its rubric result is exactly the kind of finding this index exists to publish: `runtime_governance` 2, `audit_trail` 2, `compliance` 1, and two trust-gap zeros — `outcome_settlement` and `insurance_indemnity` — the page contains no occurrence of "insur", "indemn" or "liabilit" at all.

**Two wire pieces published**, both AI agent vendors already tracked as candidates: **Neuwark** markets "outcome-based" pricing — "Pay for outcomes, not for access," and states it "sells completed claims, processed applications, closed cases—and takes responsibility for the result" — and lists SOC 2, HIPAA and GDPR, but names no insurance, liability-cap or indemnification term of its own; it also lists "Insurance" among its target customer sectors, which our editorial check confirmed is a sector label, not a coverage claim. **User Intuition**, an AI-run user-research platform, ties payment to a quality bar — "you only pay for interviews that clear the quality bar," unpaid if a transcript "misses the bar" — and states a SOC 2 Type II examination "is undergoing," not completed; it names no insurance or liability-cap terms. A third draft, on observability vendor Future AGI, was not cleared for publication this cycle.

(sources: neuwark.com; userintuition.ai — full pieces: `neuwark-outcome-based-pricing-insurance-vertical-no-coverage-disclosure`, `user-intuition-quality-gated-payment-soc2-in-progress-no-insurance-disclosure`)

## What changed in the index this week

Real counts, queried directly against `invsbcblyrjsvmygulcz` at draft time:

- **Published entities: 1,705 of 2,970 tracked** (1,264 in the candidate queue, 1 archived) — up from 1,704 of 2,965 yesterday, net **+1** published, net **+5** tracked. - **In the trailing 7 days: 102 new entities entered the index, 3,740 field values were extracted, 827 new scorecard rows were written, and 20 wire articles published** (2 of them today). - **The field-extraction number carries the same caveat as every issue since the outage began: the newest `entity_fields` row written by the automated extractor is still timestamped 2026-08-12 10:28 UTC.** The pipeline has now been down roughly five days. Every field value credited above was written before that cutoff or entered by an editor working by hand from already-captured pages — not new automated extraction. We report this plainly rather than let a rolling weekly count imply the pipeline is running. - **Rigor-equivalent coverage — 12 or more evidenced fields plus a complete, human-scored five-dimension scorecard — stands at 23 of 1,705 published entities (1.3%),** up from 21 two issues ago. Separately: **60 of 1,705 published entities (3.5%) carry at least one hand-scored (`v1`) dimension of any kind** — the wider, softer number; read the rigor-equivalent figure above for the harder bar.

## The assurance gap: three dimensions, published entities

How many published entities score 0–1 out of 3 — a trust gap under the published rubric — on the dimensions hardest to fake: a tamper-evident audit record, payment tied to a verified outcome, and insurance cover with a named carrier:

| Dimension | Hand-scored (`v1`), 0–1 of total | All rubrics, 0–1 of total scored | Two issues ago (all rubrics) | |---|---|---|---| | `outcome_settlement` | 60 of 60 (**100%**) | 165 of 202 (81.7%) | 163 of 200 (81.5%) | | `insurance_indemnity` | 58 of 60 (96.7%) | 66 of 68 (97.1%) | 64 of 66 (97.0%) | | `audit_trail` | 51 of 59 (86.4%) | 440 of 466 (94.4%) | 439 of 463 (94.8%) |

Read the hand-scored column for anything claiming a literal floor: the automated scorer (`v1-auto`), which supplies the large majority of "all rubrics" rows, still cannot emit a literal 0 on any dimension. Every one of the 60 hand-scored entities in the index scores in the trust-gap band on `outcome_settlement`, with no exception — that has now held across every issue we have published.

## One assurance/insurance development

**Munich Re aiSure's own insurance-availability field was recorded as absent in our index until this morning — and the correction is worth more than the number it changed.** The vendor's page states plainly: "**AI insurance, pioneered by Munich Re in 2018, is triggered by unexpected errors in the ongoing performance of AI models in daily business use.**" That sentence was sitting in our stored capture of `munichre.com` the whole time; a detector built this week flagged it as a likely false absence, and today's audit confirmed and fixed it, filing the field `present` rather than `no_public_information`.

We are also correcting our own prior claim about the scope of this defect, because a diligence tool that only corrects other people's numbers is not one. An earlier internal note characterized this as affecting "all four flagship insurance vendors" in the index. Today's audit checked each individually against its own capture: only Munich Re's absence was real. The other three flagged matches were pattern-matching artifacts — a navigation link containing the word "coverholder," and two carriers' own names matching a regex built to find carrier mentions. One in five of the detector's high-confidence flags on insurance and audit fields survived manual review industry-wide this run; treat the raw flag count as a review queue, not a defect tally, wherever you see it cited.

(Source: munichre.com/en/solutions/for-industry-clients/insure-ai.html, retrieved 2026-07-31.)

## Methodology

Full rubric and sourcing standard: `/methodology`. Every figure above was queried directly against the live database at draft time (August 17, 2026, 16:24 UTC) rather than carried forward from a prior issue or from another role's report.

---

*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 and cite everything. If we ever stop doing that, stop trusting the newsletter.*