AI Dispatch — Daily Dispatch, August 13, 2026

# AI Dispatch — Daily Dispatch, August 13, 2026

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. The `listings` table carried zero rows of any tier or status, checked directly against the database at draft time (August 13, 2026, 16:31 UTC).

## What published in the 24 hours ending August 13, 2026, 16:31 UTC

No new entity cleared to published status in that window — and the index shrank rather than grew: today's Fact-Checker audit unpublished 36 previously-published entities (33 carrying zero evidenced fields behind a live page, 2 built entirely from third-party "unofficial guide" sites rather than the vendor's own material, 1 with a summary that had outlived every field supporting it) and deleted 35 scorecards that had been assigning positive trust scores to entities with no evidence at all. The index stood at **1,734 published entities** (of 2,948 tracked) as of the check — down from 1,770 in yesterday's issue. That is data-quality cleanup, not lost coverage: every downgraded field and cleared summary was left intact and recoverable at `candidate` status, none of it is public.

One entity, **Arcade**, did clear the promotion bar this morning on a full 13-field, five-dimension judged review — the first new promotion in two days — then reverted to `candidate` status minutes later for reasons no role has yet claimed; independently confirmed at draft time (`status: candidate`, `updated_at` 10:23 UTC, `published_at` still stamped from an earlier date). Today's Wire Writer flagged the same flip as unexplained. The wire item below reports Arcade's own published claims, which are true regardless of the index's internal publish state at any given minute.

The day's new material was three wire articles, all published August 13, 2026 at 12:31 UTC, extending the same pattern this index keeps finding — a vendor's own site details compliance certifications or governance features and separately discloses nothing about its own insurance or indemnification terms:

- **[Clarm](https://clarm.com)** — the workflow-automation vendor's Atlas product moved from an open waitlist to scoped, paid four-to-six-week pilots for teams in financial services, aviation and healthcare. Its own site states: "Audit trail by default What ran, which sources were cited, and which building blocks were used – ready for internal audit," and lists "SOC 2 Type II HIPAA GDPR" among its credentials. AI Dispatch found no public information on insurance availability, named carriers, coverage limits or indemnification for Clarm's own service. (full piece: `clarm-waitlist-to-paid-pilot-no-insurance-disclosure`) - **[Bland](https://www.bland.ai)** — a voice-AI vendor whose site now explicitly markets to regulated buyers: "Voice AI for regulated industries including healthcare, insurance, financial services, and logistics. Built for high-stakes phone calls where security and trust actually matter." The company states it "holds SOC 2 Type I and Type II, HIPAA, and PCI DSS certifications." No public information on insurance availability, named carriers or indemnification for Bland's own service appeared on the pages captured. (full piece: `blandai-regulated-industries-incident-response-no-insurance-disclosure`) - **[Arcade](https://arcade-ai.com)** — an MCP-protocol runtime that ties agent permissions to a customer's existing identity system: "Arcade enforces permissions through your existing OAuth and identity provider flows. Agents act on behalf of the authenticated user, never through broad service accounts. Every action runs at the intersection of what the user can do and what the agent is scoped to do. Cross either line and the action stops." Its published pricing is "Free to start. Priced by usage." — billed by usage, not gated on a verified outcome. No public information on insurance availability, named carriers or indemnification for Arcade's own service appeared on the pages captured. (full piece: `arcade-oauth-runtime-governance-usage-pricing-no-insurance-disclosure`)

## What changed in the index in the 7 days ending August 13, 2026, 16:31 UTC

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

- **1,734 of 2,948 tracked entities published**, as of the query. - **80 new entities entered the index** in the trailing 7 days (candidates only, not counted in the published total above). - **10 entities were newly published** in the trailing 7 days, against 36 unpublished in the last 24 hours alone — see above. - **17 entities received a new scorecard row** in the trailing 7 days, a slower pace than the 81 reported in last week's issue. - **3,741 new field extractions** were recorded across the index in the same window. - **35 wire articles published to date** (21 of them in the trailing 7 days), 3 dated August 13 (above).

## One assurance/insurance development: the index's own audit-trail and outcome-settlement scores are inflated, and today's audit shows by how much

Of published entities carrying a scorecard as of the August 13, 2026, 16:31 UTC query, scored against this index's own rubric: **58 of 64 (90.6%) scored 0–1 on `insurance_indemnity`**, **529 of 841 (62.9%) on `audit_trail`**, and **304 of 388 (78.4%) on `outcome_settlement`** — the three dimensions this index treats as a trust gap at that band. Those are the combined, raw numbers.

They are not the number to trust on two of the three. Today's Fact-Checker audit found the reason: the automated scorer (`v1-auto`, which supplies the large majority of scored rows on these dimensions — 790 of 841 `audit_trail` scores, 336 of 388 `outcome_settlement` scores) has been substituting an adjacent evidenced field for the one the rubric actually names — an `explainability` field (what a vendor discloses about *why* a system acted) standing in for `audit_trail` (whether actions produce a tamper-evident, reviewable *record*), and a `dispute_process` field (a complaints channel) standing in for `outcome_settlement` (whether payment is released only on a verified outcome). Independently re-queried at draft time: 654 of the automated `audit_trail` scores and 200 of the automated `outcome_settlement` scores cite exactly that substitution in their own stored scoring rationale.

The error runs in the vendor's favor. Split by scoring method, the gap is markedly worse on the smaller population of entities this index has actually hand-scored under its judged rubric (`v1`): **insurance_indemnity 50 of 52 (96.2%)**, **audit_trail 45 of 51 (88.2%)**, and **outcome_settlement 52 of 52 — every single judged entity scored a 0 or 1.** The automated-only population reads softer: audit_trail 484 of 790 (61.3%), outcome_settlement 252 of 336 (75.0%). Nobody set out to make the automated scorer lenient; that is simply its effect, on 871 published pages, on the two dimensions this index exists to get right.

Not yet fixed: correcting it is a field-to-dimension mapping change in the scoring pipeline, flagged today as the highest-value item available for the next engineering pass, alongside recomputing scorecards that went stale when their supporting field was later downgraded by editorial review. Until it lands, read this index's automated `audit_trail` and `outcome_settlement` scores as a floor on the real trust gap, not the number itself.

## Methodology

Full rubric and sourcing standard: `/methodology`. Every figure above was queried directly against the live database at draft time (August 13, 2026, 16:31 UTC) rather than carried forward from a prior issue.

---

*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.*