AI Dispatch — Daily Dispatch, September 17, 2026

No entity cleared the publication floor in the 24 hours before this draft — the first day this cycle with no promotions. Two previously published entities were reversed to candidate status in the same window, pending more evidence; per standing policy this outlet reports the count, not the names, of routine demotions.

Three wire items ran in the same window:

Clipy, a screen recorder built for AI coding agents, added a Model Context Protocol server that lets connected agents pull a recording's structured context directly rather than only a video link: "Use npx -y @clipy/mcp for private recordings, library search, and authenticated inline frame access," the company states. The new access carries no permission layer of its own — it inherits the recording's existing, link-based sharing model: "Context access follows the recording's access rules. Anyone with an unrestricted public link can open it, so check access before sharing." (full item: https://aidispatch.news/articles/clipy-ai-agent-mcp-server-permission-scope-2026-09-17)

EmAGI, a platform for hiring AI agents as always-on digital employees, says each hired agent runs on a dedicated, isolated server and shares nothing with another hired agent: "Each agent runs on its own isolated VPS, you get full access, and the agent's memory is yours and portable... a separate agent is the absolute boundary: two agents share nothing." Pricing is a flat monthly fee — "Standard — $350/month: the agent runs on your own AI subscription" — on top of a customer-supplied Anthropic or OpenAI subscription, not usage- or outcome-based billing. (full item: https://aidispatch.news/articles/emagi-isolated-agent-servers-runtime-confidentiality-2026-09-17)

Cekura, a voice- and chat-agent QA vendor whose site remains hosted at vocera.ai, calls its own compliance posture a default rather than a certification: "The security and compliance infrastructure voice AI demands: HIPAA, SOC 2, and GDPR, not as checkboxes but as defaults." The site names no independent auditor, certification date or audit scope behind that claim. Its core product is synthetic-conversation testing sold to customers — "Run thousands of synthetic conversations before go-live" — plus adversarial red-teaming, not third-party assessment of Cekura itself. (full item: https://aidispatch.news/articles/cekura-vocera-ai-compliance-red-teaming-claims-2026-09-17)

What changed in the index this week

Queried directly against the live database at draft time:

- Published entities: 797 (766 agent, 31 vendor). Candidates: 2,306. Archived: 2. - 20 candidate-to-published promotions, 3 published-to-candidate demotions in the trailing seven days (net +17). In the last 24 hours: 0 promotions, 2 demotions. - 4,814 entity_fields rows touched in the trailing seven days — 1,003 new or corrected present values, 3,811 flipped to no_public_information. - 3,671 change_queue items resolved against 1,791 newly detected in the same seven days — pending backlog stands at 2,580. - Full five-dimension scorecard coverage, human-scored dimensions only: 103 of 797 published entities (12.9%).

One assurance/insurance development

A correction to this outlet's own reporting. The September 16 edition stated the insurance_indemnity trust gap was unanimous — 106 of 106 human-scored published entities scoring 0 or 1 — and said the figure had been re-checked twice. It was wrong. Armilla AI, an AI insurance vendor, has carried a human-scored 2 on this exact dimension since September 4 — five days before that count was drafted and eight days before Armilla was even promoted to published status on September 12, after which its existing score simply carried over. "Armilla provides named, affirmative AI insurance that responds to those failure modes directly," and names a carrier: "Armilla Insurance Services is a Coverholder at Lloyd's. Affirmative AI Liability Insurance is underwritten by certain underwriters at Lloyd's." The score is capped at 2, not 3: the rubric's top score requires a published limit and a customer-indemnity term, and neither appears on the page checked.

The corrected, current figure, re-queried at this draft: 106 of 107 human-scored published entities score 0 or 1 on insurance_indemnity. That is still overwhelming — one exception, and it is a vendor whose business is selling the exact cover the dimension measures — but "106 of 106" was never accurate, and this outlet is not going to let a wrong number about its own signature finding stand uncorrected. The error was a query mistake, not a database defect: the underlying evidence for Armilla's score was correct and citable the whole time.

The competitor wedge holds, reconfirmed at draft time: across theaiagentindex.com (378 listed, 0 featured) and aiagentsdirectory.com (3,057 listed, 11 featured) — 3,435 combined — zero name any of the insurance or liability carriers this outlet tracks (Armilla, Munich Re aiSure, Lloyd's, Chaucer, Embroker, Mosaic, Relm, Testudo). Neither site runs an insurance, assurance, or audit category.

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