AI Dispatch — Weekly Dispatch, August 7, 2026

# AI Dispatch — Weekly Dispatch, August 7, 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 — no `featured`/`active` listing exists in `listings`.

## What changed in the index this week

Real counts, queried at run time:

- **Published entities: 1,773** of 2,868 tracked (1,733 AI agents, 40 assurance/supplier vendors); 1,095 remain in the candidate queue awaiting verification. - **No entity has cleared to published status since Aug 5, 08:27 UTC** — roughly 56 hours flat as of this run. Daily publish volume this week by timestamp: Jul 30 — 4, Aug 1 — 889, Aug 2 — 859, Aug 3 — 0, Aug 4 — 17, Aug 5 — 3, Aug 6 — 0, Aug 7 — 0 so far. The two bulk days (Aug 1–2) account for 98% of everything published this week; growth since has effectively stopped. - **17 wire articles published to date, 3 of them today** — on Lloyd's of London, Auquan, and Vorim (below). - **Scorecard coverage: 1,155 of 1,773 published entities (65%) carry at least one trust-gap dimension; only 42 (2.4%) carry the complete five-dimension scorecard.** We report exact counts, not the total, whenever we cite a score. - **Scope note on scoring:** most scored dimensions come from `v1-auto`, an automated scorer that is structurally unable to award either a 0 or a 3 on any dimension. Only entities scored under the manual `v1` rubric can register the true floor or ceiling. Every number below that claims a literal 0 is drawn from `v1`-scored entities, not `v1-auto`.

## One assurance development: Lloyd's of London discloses no AI-specific terms at the market level

Lloyd's of London — the marketplace whose member syndicates underwrite much of the nascent AI-liability and AI-performance insurance market, including capacity distributed through Munich Re's aiSure program — publishes no public information on AI-specific audit, governance, compliance, outcome-settlement or liability terms at the market level. Lloyd's describes itself as "the only insurance marketplace of its kind in the world," bringing together "more than a hundred syndicates and thousands of investors" that "shoulder more insurance risk for every unit of capital than any other financial institution in the world," serving "200+ territories" through "50+ leading insurance companies" with "£57.9bn gross written premium." Of the 45 index fields that apply to Lloyd's as a vendor, 42 currently show no public information — every field tracking audit trail, runtime governance, compliance certification, outcome-based settlement, coverage limits and liability caps as they'd apply to AI risk specifically.

That gap is structural, not unusual: Lloyd's itself does not underwrite; individual syndicates do, and set their own terms. It means the address through which much AI insurance capacity ultimately routes discloses nothing at the market level, leaving buyers to find terms syndicate by syndicate rather than at one citable source. (source: lloyds.com; full piece: `lloyds-of-london-scale-no-ai-assurance-disclosure`)

## The assurance gap, Vol. 2: the AI insurance market's own scorecard

Two weeks ago, on four AI-insurance vendors, we found none scored above 2 on any trust-gap dimension and none cleared the gap threshold on four of five. The index now covers eight published AI-insurance and assurance vendors — Armilla AI, Chaucer Group, Coalition, Embroker, Lloyd's of London, Mosaic Insurance, Munich Re aiSure, and Vouch — all hand-scored under the manual `v1` rubric, not the automated scorer. The pattern has held exactly, at double the sample:

| Dimension | Vendors scoring 0 (of 8) | What "0" means here | |---|---|---| | `audit_trail` | **8 of 8** | No tamper-evident, customer-reviewable log published by any vendor | | `compliance` | **8 of 8** | No certification or per-engagement compliance schedule published | | `outcome_settlement` | **8 of 8** | No vendor ties payment or claims settlement to a verified outcome | | `runtime_governance` | **8 of 8** | Not natively applicable to an insurance product, scored per the published rubric regardless — every page checked describes marketing content, not enforced controls | | `insurance_indemnity` | 0 of 8 (scores run 1–2) | Their own core business — the one dimension where every vendor clears the floor, though none reaches a 3 |

**Every one of the eight vendors scores literally zero — the floor, not just "a gap" — on four of the five dimensions in a rubric they did not write and are not being singled out by.** On the fifth dimension, the one that is their actual product, none reaches full marks either: Armilla AI and Munich Re aiSure score highest at 2, for naming AI-specific coverage without publishing a limit; the other six score 1.

A concrete example of that pattern: Munich Re's aiSure program states plainly, "aiSure™-backed performance warranties enable you to indemnify your clients for their financial losses or legal liabilities directly related to AI errors." That is a real, citable indemnification claim — and it is also true, checked against the same page, that Munich Re publishes no coverage limit. That combination is not unique to Munich Re: **all eight vendors' `coverage_limits` field reads `no_public_information` — zero of eight publish a dollar limit on the AI-specific cover they market.** A buyer can find an insurer's claim to cover AI risk; a buyer cannot currently find, on any of these eight vendors' own sites, what that cover is capped at.

**Scope, stated plainly:** n=8, and every entry is an insurance or assurance vendor rather than an AI agent, so `runtime_governance` and `audit_trail` — built with agents primarily in mind — fit these vendors more loosely than they'd fit an agent. It is also not yet a claim about the wider AI agent market: the index-wide trust-gap picture (all 1,773 published entities, agents included) shows a real but structurally muddier gap — 533 of 857 scored `audit_trail` rows, 296 of 384 `outcome_settlement` rows, and 45 of 51 `insurance_indemnity` rows fall in the 0–1 trust-gap band, but the bulk of those rows are `v1-auto`-scored 1s, not confirmed 0s, because the automated scorer cannot emit a 0. The only entities we can say, today, sit at the literal floor on any dimension are the ones scored by hand — currently the eight above plus a small number of manually spot-checked others. That is a narrower, more defensible claim than "most of the index scores zero," and we are making the narrower one.

Read plainly: the industry that would price and pay out on AI risk has not yet, on its own public-facing pages, disclosed the audit trail, compliance record or outcome-linked terms that a buyer doing diligence on any AI vendor would want to see — including diligence on the insurer itself.

## Methodology

Full rubric and sourcing standard: `/methodology`. Every score above traces to one evidence record — a fetched page, a retrieval timestamp, and a verbatim quote — available on request.

---

*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.*