The wire

Screen-recorder Clipy adds MCP server for AI agents; recording access follows the same public link

The developer tool's new agent integration inherits its existing link-based sharing model, so anyone holding an unrestricted recording link can also pull the agent-readable version.

AI staffing platform EmAGI runs each hired agent on its own isolated server, blocks cross-agent memory sharing

The company says confidentiality is set per conversation and no data crosses between separately hired agents, while customers supply their own Anthropic or OpenAI subscription.

Voice-agent QA vendor Cekura calls HIPAA, SOC 2 and GDPR compliance 'defaults,' offers adversarial red-teaming

The company, whose site remains at vocera.ai, tests customers' voice agents with synthetic conversations and red-team probes; it names no independent auditor for its compliance claim.

AI Dispatch — Daily Dispatch, September 15, 2026

One flagship report and two wire items ran in the last 24 hours. The report ran our own promotion gate backwards against the 792 entities already published and found 382 (48.2%) would not pass it today, 230 of them because a quote we cite is no longer on the page it came from. The wire items — video platform Ozor and photo app FastPhoto, both newly promoted — each publish no insurance or liability terms of any kind. Separately, a field-level audit downgraded 112 rows across 65 entities, 61 of them on already-published pages, mostly nav-link and section-heading text that had been miscoded as compliance findings. Published entities: 794, net +16 over the trailing seven days.

AI agent arena LLMPvP adds MCP server, validates every move at runtime rather than trusting model output

The bring-your-own-model chess and Go arena pits agents from any provider against each other and checks each proposed move before it counts.

AI model-gateway reAPI enforces spend caps at runtime, keeps only call metadata, not content

The multi-model API aggregator says a runaway job hits a daily ceiling rather than an open bill, and that it never stores request or response content.

WordPress chat plugin YakWP keeps conversations in the site owner's own database, calls itself "GDPR Ready" without naming a certifying body

The self-hosted AI support widget routes chats through a site owner's own Gemini or OpenAI key rather than YakWP's servers, and describes its own compliance status without specifying what it was assessed against.

Video agent Ozor adds MCP access; publishes no audit trail or insurance terms

The AI video platform now lets other AI agents generate videos directly. Its public materials say nothing about runtime limits, liability or insurance.

Face-training photo app FastPhoto discloses no insurance or liability terms — not even on its data agreement page

FastPhoto trains a personal AI model on users' faces and encrypts the data. Its dedicated data processing agreement page says nothing about liability or coverage.

The Trust Gap, Vol. 7: 48% of our published entities would not pass our own promotion gate

We turned our publication standard backwards against the pages already live. 382 of 792 fail it, 230 of them because quotes we cite are no longer on the page we cite them from. Plus: the six rarest fields in a 63-field catalogue are the same six risk-transfer fields, and both of this week's apparent insurance improvements are the same artefact.

AI Dispatch — Daily Dispatch, September 14, 2026

Three wire items ran today: two newly promoted entities — a Solana trading agent (KHAN) whose ledger claims verified transactions but stops short of claiming tamper-evidence, and a workflow-automation platform (TRam Studio) that logs every run and caps spend at runtime — plus a follow-up on a still-candidate insurance vendor (MyPolicyReview.ai) that turned up an audit-trail feature this outlet had reported as absent. Across the index's 100 human-scored insurance_indemnity entities, 99 (99.0%) still sit at zero or one. Published entities: 792, net +10 over the trailing seven days.

AI Dispatch — Daily Dispatch, September 13, 2026

Three wire items ran today: a meeting-assistant app (GeekBye, still a candidate — not yet promoted) that markets itself on leaving no trace of its own use, a newly promoted site-publishing tool (Birta) that logs every publish but carries an unconfirmed "Verified by Anthropic" badge, and a newly promoted image-layer generator (Image to Layers AI) that bills only for layers actually delivered. Across the index's 95 human-scored insurance_indemnity entities, 94 (98.9%) still sit at zero or one. Published entities: 788, net +11 over the trailing seven days.

DeFi agent KHAN calls its on-chain ledger the "single source of truth," but does not claim it is tamper-evident

KHAN settles trades, payouts and treasury moves as verified Solana transactions and discloses even a losing trade, but its transparency page stops short of claiming the audit-trail properties -- tamper-evidence, hash-chaining, customer review -- this outlet's rubric scores highest.

Workflow-automation platform TRam Studio logs every run and caps spend at runtime, but leaves compliance and insurance terms undisclosed

TRam Studio pairs a per-run audit trail with enforced runtime spend limits, two trust-gap dimensions this outlet tracks, while its own compliance certification and liability terms remain unpublished.

Fuller read of AI policy-review vendor MyPolicyReview.ai turns up an audit trail this outlet had reported as absent

A follow-up extraction of MyPolicyReview.ai's site finds a lender-facing compliance dashboard with audit export -- narrowing, not closing, the gap AI Dispatch's Sept. 11 article identified.

AI-site publisher Birta claims an unconfirmed "Verified by Anthropic" badge, logs every publish

Birta hosts websites built by a user's own AI chat agent and records each publish with a timestamp; its Anthropic-verification claim has not been corroborated independently.

AI meeting assistant GeekBye markets itself as undetectable, leaves no shared record

GeekBye answers video-call questions in real time and is marketed on avoiding detection by the other party; the company says recordings stay local and are never uploaded.

Image-layer generator settles billing per output, refunds credits on failed layers

Image to Layers AI charges only for the layers it actually delivers, an example of the outcome-linked payment structure this outlet's trust-gap rubric scores separately from price.

AI insurer Armilla AI ties its warranty to Lloyd's capacity, publishes no coverage limits

The Lloyd's coverholder frames its "AI Performance Warranty" around Colorado and EU AI Act risk categories, but discloses no liability cap, dispute process or audit-trail claim.

ApexApi gateway markets autonomous crypto funding for AI agents, discloses no compliance certification

The 130-model API gateway lists "agent self-registration" and "autonomous USDC funding over x402" among its features, alongside spend caps and guardrails it says are on by default — but no audit trail, insurance or release stage.

Storm-damage research tool StormAuditor markets to insurance adjusters, discloses no compliance or audit-trail terms

The $2.99 report pulls NOAA/NWS wind and hail records for a U.S. address and states plainly it does not determine damage, causation or coverage — but names no compliance certification, audit trail or insurance of its own.

AI policy-review tool discloses its summaries are not insurance advice

MyPolicyReview.ai reads insurance policies for customers and other agents, but states its own output does not determine coverage.

Twilio's Agent Connect SDK bars HIPAA and PCI workflows outright

The middleware SDK for building LLM agents into Twilio channels holds no compliance certification and explicitly excludes two of the most regulated use cases.

Browser-automation agent Tabstack purges page data after each call, discloses no compliance certification

The hosted service, newly promoted in AI Dispatch's index, lets a caller set the scope of each task but publishes no audit trail, insurance or compliance terms.

AI Dispatch — Daily Dispatch, September 9, 2026

Two wire items ran today — Agent Zero's tool-policy update and 9bot's liability disclaimer — both entities with no compliance, audit, or insurance disclosure on record. Separately, the Fact-Checker downgraded 23 published fields where evaluation and security vendors had filed their own product pitches as third-party evaluations of themselves, and corrected eight values on Agent Zero's own record that reached beyond what their cited quote actually supports. The index moved net -1 published entity this week (10 promotions, 11 demotions), and insurance_indemnity remains at 0 or 1 for every one of the 85 published entities scored on it — a streak that has held since scoring began.

Voice AI platform Phonely runs in healthcare, insurance and financial services; vendor claims industry-standard support, publishes no certification

Wealth-management automation vendor Jiffy.ai holds five active compliance certifications and applies AI guardrails to 100% of model calls; insurance and outcome-based pricing undisclosed

LexisNexis's Protégé legal AI publishes eight compliance certifications and a responsible-AI framework; disclosed audit record limited to a 90-day window, insurance undisclosed

Agent Zero framework adds tool-policy scoping and agent-to-agent bridge; no compliance, audit or insurance disclosure found

WhatsApp automation platform 9bot disclaims 'unrestricted autonomous agent' status, caps liability broadly, offers no refund guarantee

The Trust Gap, Vol. 6: 18% of our published values no longer verify against their own source

Of 6,725 published values whose source page we have re-crawled, 1,219 carry a quote that no longer appears in the current capture. The detector we built to catch this holds 31 rows, runs green, and cannot see 89.73% of the index.

AI Dispatch — Daily Dispatch, September 7, 2026

AI Dispatch's fact-check audit cleared six candidates and promoted them into the published index today, rejecting 45 of the fields they arrived with along the way. Separately, a previously published entity was pulled back to candidate after its source page turned out to now serve a different company's product — including that company's SOC 2 claim, which had been running under the wrong name for 24 days. Three wire items ran on undisclosed insurance, audit and compliance gaps at agents newly added to the index.

AI Store Shopper browses under a named bot identity and stops before checkout by design; insurance, compliance and human oversight undisclosed

The $149 checkout-testing agent scopes itself to a merchant's public storefront and names an enforced payment stop, but discloses no compliance certification, insurance backing or human-review process.

Ad-management agent Orchestra Ads requires explicit approval before publishing any campaign; account access revocable through OAuth

Orchestra Ads ties campaign management to a business's own Google and Meta ad accounts and states it will not publish without sign-off, a governance control AI Dispatch has not found disclosed by every agent it reviews.

Voice agent vendor Ideazic puts outbound AI-calling compliance obligations on its customers, not itself

Ideazic's AI receptionist places outbound calls and escalates sensitive conversations to a human, but its own compliance note assigns consent and telecom-law obligations to the business deploying it.

Field-service AI agents from Fieldproxy run fire-safety compliance inspections behind a dollar-threshold approval gate; no compliance or insurance disclosed

Fieldproxy discloses a spend-limit approval rule and a communications log, but names no compliance certification or insurance backing for agents deployed into safety-inspection work.

AI Dispatch — Daily Dispatch, September 6, 2026

AI Dispatch's own fact-check audit scored five newly reviewed AI agents zero on insurance and indemnity today, and found two of them — an outbound sales agent and a health-care voice-agent platform — naming SOC 2, GDPR or HIPAA compliance with no certificate behind the claim. Separately, the sales agent, Artisan's Ava, cleared a field-by-field review and was promoted into the published index.

Customer-support agent YundaDesk discloses per-reply audit trace and human approval gate; names no insurance or compliance certification

YundaDesk publishes two governance controls still rare among agent vendors AI Dispatch has reviewed, alongside three undisclosed trust-gap dimensions.

Agent-hosting platform HarnessRouter names Stanford Medicine as launch partner; discloses no compliance, audit trail or insurance

HarnessRouter runs coding agents as embedded backend infrastructure with sandbox isolation as its only disclosed control.

AI Dispatch audit finds no agent among five reviewed discloses insurance or indemnity terms

A Sept. 6 fact-check across five newly reviewed AI agents found none names a carrier, a coverage limit or a customer indemnity.

Sales and voice agents cite SOC 2, HIPAA compliance; audit finds no certificate published

Two vendors name specific security standards on their sites; neither publishes a certificate, auditor or report date behind the claim.

AI Dispatch — The Assurance Gap, September 4, 2026

Insurance and indemnity: 80 published entities scored, 100% still sit at zero or one — unchanged since we started counting. But this week's sharper finding is about our own evidence, not just the vendors': 37.7% of the assurance layer's published field citations are the same sentence filed under more than one field, and on at least one page we checked, that double-counting inflated a published score, not just a field count.

Text-to-MIDI plug-in MIDI Agent discloses no audit trail, insurance or runtime limits of its own

The $49 one-time license generates melodies, chords and drum patterns from prompts routed through a user's own ChatGPT or provider account; an AI Dispatch index review found no independent usage limits, audit trail, insurance or indemnity disclosed.

AI literature-review tool Litrevu tells users to check their own school's AI policy; no compliance certification or insurance disclosed

The one-time-purchase research assistant traces citations back to a user's uploaded PDFs and offers a 30-day refund, but an AI Dispatch index review found no certification, audit trail of its own actions, or insurance published anywhere on its site.

WhatsApp automation platform WaliChat gives staff full chat-editing access; no audit trail, insurance or compliance certification disclosed

A trust-gap review of the WhatsApp agent platform, used by more than 5,000 businesses, found no disclosed insurance, audit trail or compliance certification; its one runtime-governance credit covers human staff permissions, not the AI agent's own actions.

AI reliability vendor NOFireAI caps its own liability at $100 while selling runtime governance and audit-trail assurance

The company's terms of service limit its total liability to customers to $100 in the aggregate, even as its product pitch rests on enforced runtime controls, immutable audit logs and named compliance frameworks.

AI Dispatch — Daily Dispatch, September 2, 2026

Two entities crossed the publication floor today — Tidio Copilot and Fabrile — and both immediately drew wire coverage for the same pattern: a compliance claim that names a real standard with no certificate behind it. A third wire item, on the already-published open-source framework PraisonAI, showed the opposite shape — a rare enforced runtime control sitting next to the same unpublished insurance and outcome terms found everywhere else in the index.

VoiceAIWrapper cites self-declared SOC 2 for healthcare voice deployments; no audit trail or insurance disclosed

The white-label reseller of five voice-agent platforms names a real but unverified certification and scores zero on audit trail, insurance and outcome-based payment.

Devin merges its own pull requests for Nubank; no audit trail, insurance or compliance certification disclosed

Cognition Labs' coding agent ships PRs after picking up CI and review feedback, with a human approving changes but no governance, audit, compliance or insurance published.

Autonomous sales agent Gojiberry AI can run outreach unsupervised; scores zero on audit, compliance and insurance

The prospecting agent emails and messages leads for 2,000+ teams with an optional, switchable human review step and no disclosed audit trail, insurance or outcome-linked pricing.

Free Tidio Copilot extension runs on Claude for 300,000+ businesses; GDPR and CCPA compliance self-declared, no audit trail or insurance disclosed

The Chrome extension drafts support replies platform-wide. Compliance claims carry no auditor or certificate; runtime governance, audit trail and insurance are all unpublished.

Fabrile calls itself "ISO 42001 aligned" and "EU AI Act ready"; neither is a certification, and no audit trail or insurance is disclosed

The customer-support agent names a real standard and a real statute without the certificate, auditor or audit date those claims require.

Open-source PraisonAI enforces fail-closed prompt-injection scanning and logs tool calls; compliance, insurance and outcome-linked payment remain unpublished

A rare enforced runtime control and a real tool-call audit log sit alongside the same absence of certification, cover and settlement terms found industry-wide.

AI Dispatch — Daily Dispatch, August 31, 2026

Four entities clear the Fact-Checker's field-by-field remediation and return to the published index -- WaliChat, VoiceAIWrapper, Devin and ZoomMate, all scoring zero on insurance cover; three wire items detail a bank-deployed coding agent, a compliance-tools vendor selling to insurers while carrying none of its own cover, and a no-code builder that stores customers' AI-provider credentials in its hosted VM.

The Trust Gap, Vol. 5: the absence rate fell 6.5 points and none of it was new evidence

A publication floor removed 948 thin entities from our index and every aggregate improved. Decomposed against the same population, the week actually ran against us. What survives the population change: 162.6 control disclosures for every risk-transfer disclosure, measured on identical denominators.

Workflow agent Lindy publishes approval gates for outside-facing actions; SOC 2, GDPR and HIPAA remain self-declared, no insurance disclosed

The Slack, iMessage and browser-based automation agent says nothing irreversible happens without a named approver and scopes its own channel visibility; SOC 2 Type II, GDPR, HIPAA and PIPEDA compliance is asserted on the vendor's own page with no certificate, auditor or report date, and no insurance or liability-cover term is published.

AI Dispatch — Daily Dispatch, August 30, 2026

Five entities clear the Fact-Checker's field-by-field remediation and return to the published index, eight days after the publication floor demoted them; three wire items -- a web-scraper serving 5 billion requests, a lender-underwriting tool, and a HIPAA/GDPR/SOC2 "aligned" compliance line -- catalog the same zero-insurance pattern once more.

Coding agent Devin cuts bank migration times to weeks at Nubank; discloses no audit trail, compliance certification or insurance terms

The vendor says Nubank business units completed migrations in weeks rather than months or years using Devin, and doubled task-completion scores after fine-tuning on the bank's own examples; no compliance, audit-trail or insurance disclosure appears on the page reviewed.

AI vendor Kolena sells "audit-ready" compliance tools to insurers; discloses no insurance or liability cover of its own

Kolena markets SOC 2 and HIPAA compliance and 99%-plus extraction accuracy for insurance underwriting and claims work; the company publishes no carrier, coverage limit or indemnity for its own product.

No-code app builder AppWizzy stores customer AI-provider credentials in a hosted dev environment; discloses no insurance or independent audit

AppWizzy says more than 1,275 companies use its AI app-building platform, including in healthcare; provider credentials may be stored inside its hosted virtual machine, and the company discloses no insurance, liability cap or third-party audit.

Web-scraping agent Firecrawl serves 5 billion requests for AI workflows; discloses no governance, audit trail or insurance terms

The company reports P95 latency of 3.4 seconds across millions of searches and scrapes. An AI Dispatch review found no published evidence on any of the five trust-gap dimensions it uses to assess agentic AI vendors.

AI underwriting tool Vision prices Amazon acquisition reports at $149; discloses no insurance or liability terms for lender use

Zsell.ai markets the tool to lenders and investors underwriting FBA acquisitions and working-capital loans. It does not disclose cover, an audit trail or certification.

AI docs agent AUM markets itself to law firms and hospitals as "HIPAA, GDPR, SOC2 aligned"; no certification is published

AUM sells itself into legal, healthcare, financial and government work on the strength of a compliance claim it does not back with an auditor, a certificate or a report.

Data agent Dot cites customer ROI as high as 18x in vendor-published case studies; discloses no insurance or liability-cap terms

Named analysts at Choco, Duolingo, Emerge and KRY are credited with the figures in case studies Dot publishes about itself; the company states no insurance, carrier or liability-cap terms of its own.

Solana launch agent Fixr says it never holds customer funds; discloses no liability cover or performance data

Fixr mints vanity Solana token addresses for a flat 0.001 USDC fee and hands the signing key back to the customer, but publishes nothing on insurance or independently verified performance.

Apify Store claims SOC 2, GDPR and CCPA compliance and 99.95% uptime on its own site; discloses no auditor, insurance or liability terms

The marketplace of more than 58,000 AI automation tools states its compliance and uptime figures without naming an auditor, and publishes no insurance carrier, liability cap or indemnification term.

Cursor claims half the Fortune 500, Agentforce 18,000 companies; neither discloses an audit trail or AI liability terms

The coding agent and Salesforce's enterprise platform cleared field audits and joined the published index this week; between them, human oversight is a post-hoc review or an escalation fallback, and no runtime governance, certification or insurance is disclosed.

Workflow agent Lindy requires a named approver before irreversible actions; discloses no insurance or audit-log retention terms

The vendor's site describes Slack-channel scoping and a required human sign-off unusually specific for the category, then publishes no liability cap, indemnification term or audit-trail retention policy.

Voice agent Outcall AI runs healthcare and debt-collection calls on proprietary models; discloses no human oversight or insurance terms

The company says its system handles patient and creditor calls around the clock and states a HIPAA compliance commitment, but the site checked publishes no runtime governance, audit trail or AI liability coverage.

AI payroll agent Deel moves customer funds for 40,000-plus companies; discloses no insurance, audit trail or governance terms

The platform's "Actionable AI" approves hiring, payroll and IT workflows and lets customers transfer money to any bank in the world, but publishes nothing on liability, indemnification or runtime controls.

AI Dispatch — Daily Dispatch, August 25, 2026

Two wire items, one flagship self-audit, and the insurance dimension still stuck at zero.

AI testing agent KushoAI can block customer software releases; company discloses no insurance or liability terms

KushoAI's platform acts as an "AI gatekeeper" inside customers' CI pipelines, citing SOC 2 and ISO compliance and third-party penetration testing; it publishes no insurance, indemnification, liability cap or dispute process for that gating authority.

AI video-agent vendor HeyGen cites GDPR, SOC 2 and EU AI Act in compliance list; discloses no insurance or liability cover

HeyGen names five compliance frameworks in a single line on its security page and says it logs user and admin activity; it publishes no insurance, indemnification, liability cap or runtime-governance terms for its AI video-generation product.

Q&A platform AgoraDigest has AI agents challenge each other's answers; discloses no insurance, compliance or evaluation data

AgoraDigest synthesizes answers from a roster of independent AI agents into a versioned, citable record that humans can vote on and agents can challenge; the company publishes no third-party evaluation, compliance certification, insurance or liability information.

AI Dispatch — Daily Dispatch, August 24, 2026

Eight entities cleared the promotion gate for the first time in five days, including the index's first-ever audit_trail score of 3; insurance and indemnity stays the emptiest dimension measured.

The Trust Gap, Vol. 4: our own liability field was 87.5% wrong, and the absence rate did not actually improve

The headline absence rate fell from 77.90% to 73.37% this week. We changed the denominator; the industry did not move. Plus a complete census of our liability_cap field, seven of whose eight values turned out to be misfiled — in our favour.

Stripe's Agentic Commerce Protocol lets AI agents spend user funds; company discloses no liability cover for agent transactions

Stripe processed $1.9 trillion in payments in 2025 and now lets AI agents earn and spend funds on a user's behalf through its agent toolkit and Agentic Commerce Protocol; its site publishes extensive general security controls but no insurance, liability cap or indemnification specific to agent-initiated transactions.

AI agent observability vendor LangWatch discloses no insurance or liability cover for its own product

LangWatch sells the audit logs, evaluation scoring and red-teaming that other companies use to govern their AI agents; its site names an ISO 27001 certification and a compliance-monitoring vendor but publishes no insurance, liability cap, or third-party evaluation of its own claims.

Agent-identity vendor ProofChain markets audit trail to EU AI Act Annex IV; insurance and indemnification undisclosed

The blockchain-based product says its encrypted audit logs map directly to the EU AI Act's Annex IV technical-documentation requirement, but its page publishes no insurance, indemnification or independent compliance verification.

Legal AI vendor GenieAI claims 90% accuracy beating Claude Cowork, ChatGPT in self-run benchmark; no independent verification

Genie's site names two competitor products by name in a self-administered accuracy comparison; the company publishes no benchmark methodology, third-party evaluation, or insurance and audit-trail terms for a product used in legal contract negotiation.

MSP automation vendor Rallied lets customers set autonomous agent scope; audit trail and insurance undisclosed

The IT-helpdesk agent, sold to managed service providers with standing access to reset passwords and modify identity systems, states SOC 2 compliance without a published audit report and discloses no audit-trail or insurance terms.

AI Dispatch — Daily Dispatch, August 22, 2026

Three wire pieces on insurance and compliance claims that name no carrier, limit, or process; a live count of self-corrections and the still-empty insurance_indemnity field set; and Munich Re's aiSure warranty reaching two new distribution partners with no published claims mechanics.

AI email agent vendor Virtualworkforce.ai claims ISO 9001, 27001 and NIS2 certification without published audit

The vendor, whose agents draft email replies and write data into ERP systems from Outlook and Gmail, names three security certifications on its marketing page but publishes no certificate, auditor or scope for any of them.

AI hiring platform RightMatch AI displays SOC 2 badge without published audit report

The recruiting agent, which says it has screened more than 600,000 resumes and now runs inside ChatGPT, states its product meets GDPR, EU AI Act and NYC bias-audit requirements — without publishing the audits, certificate or auditor behind any of the claims.

AI pentesting vendor Agent Shield discloses liability insurance, publishes no carrier or coverage limit

The autonomous-agent security tester, sold under the brand Tough Love Security into HIPAA-regulated medical practices, states it carries professional liability cover but names no insurer, limit or claims process.

Memory-layer vendor Mem0 claims SOC 2, HIPAA and GDPR compliance; insurance and liability terms undisclosed

Mem0, an AI agent memory infrastructure product used by more than 90,000 developers, logs every read and write and cites three compliance frameworks — but publishes nothing on insurance, liability caps or dispute resolution.

Munich Re extends AI performance-warranty product through Mosaic, iTOO; claims mechanics undisclosed

aiSure, Munich Re's AI-error warranty product, now reaches South Africa and is distributed through Mosaic Insurance — but the reinsurer publishes no coverage limit or claims process for either arrangement.

GitHub automation agent Maige can act with full UI-level access; site discloses no permission scopes or incident-reporting process

Maige embeds a customer's entire codebase on setup and can take any action available in GitHub's interface, according to its own site. It names no runtime-governance framework, compliance certification or incident-reporting process.

Linear centralizes MCP agent access through Okta identity; site names no liability cap or SLA terms

Admins can now manage which agents and employees reach Linear through the Model Context Protocol centrally, tied to Okta identity, per a changelog reviewed by AI Dispatch. Linear's data terms disclose SOC 2 Type II certification and customer audit rights, but no liability cap, SLA or third-party evaluation.

LM Studio adds agent safety controls in latest release; company publishes no compliance certification or liability terms

The local AI runtime's changelog shows new permission prompts before destructive file actions and clearer OAuth authentication for external tool servers. LM Studio's public pages name no compliance certification, service-level agreement or liability cap.

Superagent positions itself as security layer for AI-native developer teams

The company's site describes runtime guardrails and self-serve red-teaming for coding agents, with no published audit-trail, compliance, insurance or settlement terms.

Hiscox to acquire Corix, the underwriting arm behind insurance broker Vouch

The deal expands underwriting capacity for Vouch, one of the brokers AI Dispatch tracks in the market for AI-related liability coverage.

Enterprise agent platform Cortx runs in finance and government today; ISO and SOC 2 claims are self-declared

Cortx, from XNODE Inc., is live today at hedge funds and government agencies in the US and India and disclaims liability for outputs from the third-party AI models it uses without naming them.

Insurance broker Vouch discloses no AI-specific coverage terms; site does not say what cover it can place

Vouch markets itself as 'the insurance broker for leaders' to technology and financial-services clients, but its site names no AI-specific coverage, carrier or liability terms and does not say whether it places AI liability cover at all.

Recruitment AI platform Skima cites EU, NYC bias-audit alignment; no certificate or audit result is published

Skima AI screens job candidates from 15 connected hiring systems and says its scoring is 'auditable,' but its EU AI Act and NYC Local Law 144 alignment claims, and its SOC 2 and GDPR badges, are self-declared with no certificate or audit result published.

AI Dispatch — Daily Dispatch, August 17, 2026

No first-time promotions today, but two entities restored after real fixes, two wire pieces on outcome-based and quality-gated vendors, and Munich Re aiSure's own insurance field corrected — plus a correction to our own overstated prior claim about it.

The Trust Gap, Vol. 3: we audited our own absences, and the correction runs against us everywhere except insurance

77.90% of our published data points say no public information. This volume measures how many of those are wrong — by hand, row by row — and reports the answer even where it damages our own thesis.

LLM observability vendor KeywordsAI traces every model call and blocks spend at set limits; its site names no insurance or outcome-based settlement terms

KeywordsAI blocks spend before budgets run out and traces every model call, but its ISO 27001, SOC 2 and GDPR claims are self-declared and its site names no insurance or outcome-based settlement term.

AI Dispatch — Daily Dispatch, August 16, 2026

Two entities cleared to published status, three wire pieces on already-tracked candidates, and a four-day extraction outage the index is reporting on itself rather than hiding.

AI agent vendor Neuwark markets outcome-based pricing and lists insurance among its target industries; its site discloses no insurance, liability-cap or indemnification terms of its own

Neuwark says it charges for completed outcomes, not access, and logs every agent action — but names no insurance carrier or liability cap for when those outcomes go wrong.

AI research-agent vendor User Intuition ties payment to a quality bar and cites a SOC 2 Type II exam underway; its site names no insurance or liability-cap terms

User Intuition says it only charges for interviews that clear a quality threshold, but discloses no insurance, liability cap or indemnification term for when a finding is wrong.

AI Dispatch — Daily Dispatch, August 15, 2026

Saturday digest: three wire pieces on already-tracked vendors, a fabricated-pricing catch and an 18-entity identity-mismatch sweep, and where the assurance-gap numbers moved after three new full-audit promotions.

Web-agent vendor TinyFish publishes an open, 300-run Mind2Web benchmark score; its page names no compliance certification or insurance terms

TinyFish, which builds infrastructure for AI agents that act on live websites, says its agent's 89.9% Mind2Web benchmark result is drawn from 300 runs the company has open-sourced for independent checking. Its site names no compliance certification, audit trail or insurance term.

Momentic AI backs its browser test-automation agent with a SOC 2 audit and a contractual 99.99% uptime SLA; its page names no insurance or indemnity terms

Momentic AI, whose agents write and run automated software tests, publishes enterprise access controls and a SOC 2 compliance claim alongside a contractually backed uptime guarantee. The company's site names no insurance carrier, liability cap or outcome-based pricing.

Ad-creative agent Segwise cites SOC 2 Type II compliance and per-customer data isolation; its page names no insurance or audit-trail terms

Segwise, whose agent generates and revises advertising creative on its own, says customer data stays isolated per workspace and that it holds SOC 2 Type II compliance. Its published pricing is trial-based, not tied to a verified outcome, and its site names no insurance carrier or audit-trail process.

CodeRabbit adds regional CLI auth, a branch-commit option for Fix CI and Jira context exclusions; changelog names no insurance terms

Three dated August changes to the AI code-review platform touch authentication, automated CI fixes and workspace context controls.

AI Dispatch — Weekly Dispatch, August 14, 2026

Fact-Checker pulled 18 published entities for brand misattribution and closed the two-day "arcade" revert mystery (a daily local import job silently overwriting entity status). Separately: the corrected trust-gap scorer shows the audit_trail gap is worse than yesterday's buggy number, not better — 94.6% of scored published entities land in the 0-1 trust-gap band, up from 62.9% under the mapping bug.

ElevenLabs opens Dubbing v2 to its API and adds guardrail-tracking fields to ElevenAgents conversation logs; changelog names no insurance terms

An Aug. 10 update lets developers filter conversations by which automated guardrail fired, and exposes a project-based dubbing workflow via API.

LM Studio's Bionic agent adds sandboxed cryptography operations, safer shell-command reviews and file-deletion prompts; changelog names no insurance terms

Two dated August releases add security-relevant guardrails to LM Studio's local coding-and-automation agent.

AI Dispatch — Daily Dispatch, August 13, 2026

Zero new entities published today — the index shrank by 36 as Fact-Checker cleared zero-evidence publishes; today's audit also found the automated scorer inflating audit_trail and outcome_settlement scores across 871 published pages.

Linear adds a GitHub Copilot cloud-agent handoff and signed commits for coding sessions; changelog names no insurance or indemnity terms

Linear's July 30 changelog lets GitHub Copilot users route issues to Copilot's cloud agent from inside Linear and adds SSH/GPG-signed commits for its own agentic coding sessions, plus general availability for AI-generated pull-request review summaries. The changelog entry names no insurance, indemnity or liability terms.

AssemblyAI sets Universal-3.5 Pro as its default speech model Sept. 2, retires Universal-3 Pro; changelog names no insurance terms

AssemblyAI said accounts that do not pin a specific speech-to-text model will be automatically routed to its new flagship model, Universal-3.5 Pro, starting September 2, and that the older Universal-3 Pro models are being retired. The changelog entry does not name an insurance carrier, coverage limit or indemnity term.

Agent-sandbox vendor E2B retires self-hosted reverse-proxy service, repairs a broken SSL trust store in template builds; changelog names no insurance or indemnity terms

E2B, which runs isolated cloud sandboxes for AI coding and computer-use agents, said in an August 10 changelog that self-hosted clusters no longer ship a reverse-proxy component and that a bug leaving Java trust stores silently broken inside built templates has been fixed. The entry does not name an insurance carrier, coverage limit or indemnity term.

Voice AI vendor Bland markets itself for healthcare, insurance and financial-services calls, publishes 24/7 incident response, discloses no insurance terms of its own

Bland's website states its voice agents are built for regulated industries and describes a documented incident-response process with SLA-bound notifications. The company publishes no insurance coverage, indemnification or outcome-linked pricing for its own service.

AI workflow vendor Clarm moves from waitlist to paid pilots, cites SOC 2 Type II, HIPAA and GDPR, discloses no insurance terms of its own

Clarm's Atlas product has moved from a public waitlist to scoped, paid pilots for teams in financial services, healthcare and aviation. The vendor publishes audit-trail and compliance claims but no insurance, indemnification or outcome-linked pricing terms.

Agent tool-runtime Arcade enforces permissions through customers' own OAuth flows; its pricing is usage-based, not outcome-linked, and it discloses no insurance terms

Arcade, a Model Context Protocol runtime for connecting AI agents to enterprise systems, ties agent permissions to a company's existing identity provider and logs actions for audit. Its published pricing is billed by usage rather than a verified outcome, and the vendor names no insurance, indemnity or liability terms.

DeepOpinion rebrands as Otera, keeps insurance and finance workflow-automation product, discloses no insurance terms of its own

The enterprise AI-agent vendor is now presented entirely under a new name on its own site. The underlying product, and its lack of published insurance or liability terms, is unchanged.

AI governance vendor Credo AI cites Forrester Wave leader placement, sells into insurance and healthcare, discloses no insurance terms of its own

Credo AI is one of a small number of vendors in AI Dispatch's index whose evaluation claim traces to a named third party rather than a self-issued badge. It publishes no insurance or indemnification terms for its own platform.

Voice AI vendor Speechmatics cites independent Pipecat benchmark for lowest word-error rate, discloses no insurance terms of its own

Speechmatics points to a third-party benchmark rather than a self-reported figure for its core accuracy claim. Runtime governance, insurance and indemnification terms are unpublished.

The Trust Gap, Vol. 2: agents disclose how they are governed, not who pays when they fail

Across 1,785 published AI agents and assurance vendors, 77.66% of what we tried to learn is not public. Controlling for how much each company documents, operational governance is disclosed slightly above average and risk transfer at one sixty-seventh of it. We also hand-audited our own insurance data and 21 of 31 rows failed.

AI skills platform Agentman says Medman cuts eligibility-check costs 94%, discloses no insurance terms of its own

The company's flagship deployment logs an audit trail across more than 1,300 payers and claims a $7.97-to-$0.50 drop in per-check cost; its site names no insurer, indemnification policy or liability cap for the product.

Financial-services agent platform Gradient Labs runs disputes and insurance claims to FCA, EU AI Act standards, discloses no insurance terms of its own

The company says its agents resolve chargebacks and file insurance claims under SOC 2 Type 2 certification with human sign-off before escalated decisions; its own site lists no insurance, indemnification or liability-cap terms.

Microsoft says Copilot Cowork is used by more than half of Fortune 500, discloses no indemnification terms specific to the product

Cowork pauses for approval on sensitive actions and logs what Microsoft calls an auditable record; the company's product page names no insurance, indemnification or liability-cap terms of its own.

AI Dispatch — Daily Dispatch, August 10, 2026

Three wire items published August 10, 2026 examined vendors selling governance, audit and automation into regulated industries while disclosing nothing about their own insurance or liability terms. No new entities published in the same 24 hours; the index held at 1,785.

AI platform DeepOpinion processes 600,000 regulated cases a year, discloses no insurance or liability terms of its own

The vendor cites SOC 2, ISO 27001 and EU AI Act compliance for automating insurance claims and loans. Its own coverage, indemnity and dispute-resolution terms are not published.

AI governance vendor Fiddler enforces runtime guardrails for banks and insurers, discloses no insurance terms for its own product

Fiddler AI says its guardrails block policy violations in under 80 milliseconds and log every enforcement decision. Coverage, indemnity and liability terms for the platform itself are not published.

Open-source coding agent Kortix Suna gates merges behind explicit permission grants, discloses no insurance terms

The agent runs unattended on cron schedules and webhooks but requires a human-approved config change before it can widen its own access. SOC 2 audits are in progress, not complete.

Confidential-computing vendor Phala cryptographically attests what ran on its infrastructure, discloses no insurance terms

Phala issues verifiable runtime proofs for AI workloads processing financial and health data, and lists SOC 2 and HIPAA compliance, but its site names no insurance, indemnification or liability terms.

Data-agent platform Wren AI advertises full query-level audit logs, discloses no insurance or compliance-certification terms

Wren AI cites banking and healthcare deployments and a row-level audit trail "enforced at query time," but names no SOC 2, HIPAA or equivalent certification, and no insurance terms, for its own platform.

Voice AI platform Bland reports 591 million calls resolved, discloses no insurance or liability terms for its own service

Bland markets its agents into healthcare, insurance and financial services and lists SOC 2, HIPAA and PCI DSS certifications, but publishes no insurance, indemnification or dispute-resolution terms of its own.

AI Dispatch — Weekly Dispatch, August 7, 2026

The AI insurance market scores itself: eight published vendors, evaluated by hand, score zero on four of five trust-gap dimensions -- and none publishes a coverage limit.

Insurance-industry AI vendor Indemn cites NAIC-aligned design and 90% autonomous binding, discloses no indemnification or coverage terms for its own service

Indemn markets audit trails and human-in-the-loop escalation to insurance carriers while publishing no liability cap, insurance or dispute process for its own AI.

AI call-QA vendor Voxjar cites 1,400+ business customers and SOC 2 certification, discloses no insurance or audit-trail terms

Voxjar publishes a security attestation and customer count but no liability coverage, indemnification or dispute process for its AI scoring.

AI research tool THEO offers a free re-run for false claims, discloses no insurance, audit-trail or SLA-remedy terms

THEO markets verified, cited sourcing as its core promise but publishes no defined remedy beyond a vendor-discretion re-run.

AI security vendor Vorim pairs runtime governance and tamper-evident logging with no insurance disclosure

Vorim AI markets permission controls and an audit trail for regulated industries. Its site publishes nothing on insurance, liability caps or dispute resolution.

Financial-research agent Auquan says 40% of top 50 global firms use its platform, discloses no insurance terms

Auquan reports SOC 2 and ISO 27001 certification and a 2025 Gartner vendor citation for its institutional-finance agents. It publishes nothing on insurance, liability caps or outcome-based pricing.

Lloyd's of London, marketplace underpinning much of AI risk coverage, discloses no AI-specific terms at the market level

Lloyd's reports £57.9 billion in gross written premium across its syndicates. AI Dispatch's index finds no public information on audit trail, governance or liability terms specific to AI risk.

Observability platform Phoenix discloses audit-trail tracing, silent on compliance certification and insurance

Arize AI's open-source tracing tool markets itself as an audit layer for AI applications but publishes no compliance certification, liability terms or insurance coverage of its own.

Desktop AI copilot XXAI discloses pricing and model-aggregation setup, no public information across 17 assurance fields

The Mac and Windows overlay app aggregates third-party AI models and states it does not train on user data, but publishes nothing on governance, audit trail, compliance or insurance.

Sales-call agent Insyghtful.ai can act mid-call without a human prompt, discloses no audit trail or liability terms

The beta product flags slipping deals in real time and lets a rep act immediately; its public materials say nothing about oversight, governance limits or insurance.

ElevenLabs Agents discloses audit logging and compliance certifications, no insurance or outcome-settlement terms

The voice AI agent platform names six compliance certifications and describes workflow-gate governance and action logging, but its product page carries no information on insurance, indemnification or outcome-based settlement, AI Dispatch's review found.

Three AI tools added to index this week disclose no audit trail, compliance certification or insurance terms

A video generator, a small-business assistant and a developer-tools site joined AI Dispatch's index in the past day; only one makes any public governance claim, and none discloses audit, compliance or insurance information.

Multi-model chat platform BrainChat discloses runtime spend caps and compliance certification, rare among tracked AI vendors

The browser-based workspace's site states SOC 2 Type II and GDPR compliance and admin-enforced per-seat budgets — evidence on two of five trust-gap dimensions where most vendors in AI Dispatch's index publish nothing.

AI video-generation aggregator VideoWeb AI joins index disclosing no governance, audit or insurance evidence

The service, which routes user requests across Kling, Runway, Luma and other third-party video models, publishes pricing and privacy claims but nothing on any of AI Dispatch's five trust-gap dimensions.

AI agent platform processes 600,000 insurance claims a year, discloses no insurance terms of its own

DeepOpinion's Otera platform automates claims and policy workflows for Fortune 500 insurers and publishes audit-trail and compliance claims. On insurance and outcome settlement, the two trust-gap dimensions closest to its own beat, it publishes nothing.

AI penetration-testing vendor pairs its own liability insurance with a money-back delivery guarantee

Agent shield, marketed as Tough Love Security, is the only entity among the 20 AI Dispatch has fully audited so far to show evidence on both the insurance and outcome-settlement dimensions of its trust-gap rubric at once.

Terminal AI agent gptme requires confirmation by default, ships opt-in autonomous mode

The open-source, provider-agnostic coding agent gates tool actions behind human approval unless a flag turns that off, according to documentation newly added to AI Dispatch's index.

Nine AI insurance vendors score zero on governance, audit and outcome evidence, index finds

Each publishes a description of what it covers. None publishes evidence tying that coverage to a runtime control, an audit trail, a compliance schedule or an outcome-based payout, according to AI Dispatch's trust-gap scorecards.

AI governance platform becomes first in AI Dispatch's index to hit top score for runtime enforcement

Holistic AI's "Guardian Agents" earn a 3 of 3 on the runtime-governance trust-gap dimension; rival Credo AI scores 2 for describing similar controls in less specific terms.

Four AI insurers describe their coverage. None publishes a limit.

A review of the AI liability insurance market finds a common pattern: affirmative coverage language, no public dollar figures.

In AI insurance, buyers are dealing with two different kinds of company

Some AI insurance vendors write and hold the risk. Others place it with someone else. The distinction is not always obvious from the marketing.