Quick answer: On our eight-gate regulated-fit test, Sierra scores 15 of 16 and Lorikeet 14. Sierra leads on in-conversation card payments (PCI DSS Level 1 service provider) and named hosting regions, and adds FedRAMP High and AIUC-1. Lorikeet leads on a public BAA commitment and published prices, and names data access logs and consent reports. As at 4 October 2026.
Sierra and Lorikeet both sell AI agents that resolve customer conversations end to end across voice, chat, SMS and email, and both pitch regulated buyers directly: Sierra through financial services (banking, lending, insurance, wealth) and healthcare (payers, providers, revenue cycle management, pharmacy and PBMs, and medical devices), Lorikeet through separate financial services, healthcare and insurance pages. Both run a Vanta-hosted trust centre (Sierra, Lorikeet) and both list SOC 2, ISO 27001:2022, HIPAA and GDPR. The real question is which remaining controls your regulator, Qualified Security Assessor (QSA) or privacy officer will insist on. Every vendor fact below comes from that vendor’s own pages, read on 4 October 2026. Lorikeet’s ranked guide for lenders is used only for what it says about Lorikeet itself; neither vendor’s claims about the other were used, and we name no customer of either.
Sierra vs Lorikeet: which handles regulated industries better?
Sierra edges it on our scoring, but not across the board. Scored on eight gates a regulated security review walks through, Sierra lands on 15 of 16 and Lorikeet on 14 of 16, and Lorikeet leads on the one gate Sierra drops, health data. A one-point gap still leaves the decision to the gates your compliance function will not waive.
| Control a regulated buyer asks about | Sierra (as at 4 October 2026) | Lorikeet (as at 4 October 2026) |
|---|---|---|
| Regulated work its own pages name | Card and account support, fraud and dispute resolution, loan pre-qualification, application updates, quote and bind, policy servicing, first notice of loss (financial services page); triage and routing, selection and onboarding, prior authorisation, claims support, payments and billing (healthcare page) | Disputes, chargebacks, loan and account servicing, payments, identity verification (financial services page); prescriptions, prior authorisations, benefits verification, insurance claims (healthcare page); FNOL, claims status, policy servicing, premiums (insurance page) |
| Attestations listed on its trust centre | ISO/IEC 27001:2022, ISO/IEC 42001:2023, SOC 2, HIPAA, PCI DSS 4.0.1, GDPR, CCPA, CSA STAR, EU AI Act, FedRAMP High, AIUC-1 | Trust page: SOC 2 Type II, ISO 27001:2022, HIPAA, GDPR; trust centre compliance list shows ISO 27001:2022, SOC 2 and HIPAA, with a GDPR attestation report among its resources |
| Card payments | PCI DSS Level 1 Service Provider; card data captured by keypad (voice) or embedded form (chat) and routed to your processor; agent sees status and last four digits only | No PCI DSS attestation; FAQ says it does not store or process raw card data and works through processors’ secure tokens |
| Health data | HIPAA listed; trust centre holds a 2026 HIPAA audit report (request access); business associate agreement terms not stated on the public pages we read | States it signs Business Associate Agreements; PHI handled on a minimum-necessary basis with automatic redaction |
| Guardrails | Supervisor models screen input and audit every response; deterministic guards (for example, account tools unavailable until the customer is authenticated); off-limits topic filters | Deterministic boundaries enforced in code (customer isolation, server-side identity validation, scoped tool access, execution caps); prompt-injection classifier; response checks that block, rewrite or escalate |
| Audit evidence | Agent Traces for every agent message; release governance with a Reviewer role, line-by-line diffs and immutable release snapshots; conversation logs exportable through export APIs, and agent code in a Git repository with change history (“Your agent, laid bare”) | Audit trail exports, data access logs, consent reports and incident reports for regulatory reporting (FAQ); configuration changes through its MCP server logged with a full audit trail |
| Human handoff | Routes to the right team member with an automatically generated summary (Agent SDK page); voice escalates with full context and history; Live Assist for human reps | Escalates with complete history and data, routes by issue type with an audit trail of why, escalations not charged; Resolution Loop “Take Over” mode |
| Where data is stored | Subprocessor list: AWS “localization available in United States, European Union, Southeast Asia, Japan, Australia” | Google Cloud private VPC, US primary hosting, storage residency in Australia and the EU (trust page); pricing table lists USA for Start and Scale, Custom for Signature |
| Where models run | AWS (which includes AI services) lists Australia among its localisation options; 9 other AI subprocessors list US and EU localisation; Anthropic lists the United States only | Every model-inference subprocessor that shows a location shows the USA; zero-data-retention terms with all model vendors |
| Deployment model | Hosted platform; FedRAMP High; built with Agent Studio (no code), Agent SDK (code) and Sierra’s agent development team | Hosted on Google Cloud, no public internet access to production, row-level-security tenant isolation; plugs into your existing helpdesk with forward-deployed support |
| Published prices | None; outcome-based pricing agreed per customer (sierra.ai/pricing returns a 404) | Start $2,100 a month, Scale $5,100 a month (paid annually, USD); Signature custom with a $75K platform fee |
The eight-gate regulated-fit test, scored
We call this the Eight-Gate Regulated-Fit Test. Each gate is one question a regulated security review asks, scored against the vendor’s own public pages: 2 means a page states the control with specifics; 1 means it is stated only in part, only on request, or the vendor’s pages disagree; 0 means it was not found on the pages we walked. The 1s are where your questions should go.
| Gate | Sierra | Why | Lorikeet | Why |
|---|---|---|---|---|
| 1. Attestations listed | 2 | Eleven items on its trust centre, including ISO 42001, FedRAMP High and AIUC-1 | 2 | SOC 2 Type II, ISO 27001:2022, HIPAA and GDPR, reports downloadable under NDA |
| 2. Card data | 2 | PCI DSS Level 1 Service Provider; Attestation of Compliance and shared responsibility matrix in its trust centre | 1 | Documents a token-only pattern but says it “does not hold PCI DSS certification today” |
| 3. Health data | 1 | HIPAA listed with a 2026 audit report, but no public statement on signing BAAs | 2 | Publicly states it signs BAAs; minimum-necessary PHI handling |
| 4. Guardrails | 2 | Supervisors plus deterministic guards, described layer by layer | 2 | Code-enforced boundaries plus AI-layer checks, described tier by tier |
| 5. Audit evidence | 2 | Traces on every message, Reviewer sign-off with diffs, immutable releases, Git change history, and export APIs for conversation logs | 2 | Names audit trail exports, data access logs, consent reports and incident reports for regulatory reporting |
| 6. Human handoff | 2 | Summary on handoff; full context on voice; Live Assist | 2 | Full history, issue-type routing, audit trail of why |
| 7. Residency and model location | 2 | Names storage regions including Australia, and a location for every subprocessor | 1 | Australian and EU storage on the trust page, but its pricing table lists USA on both published plans |
| 8. Deployment model | 2 | Hosted, FedRAMP High, three build routes described | 2 | Hosted, private VPC, tenant isolation described |
| Total | 15 of 16 | Loses a point on health data | 14 of 16 | Loses a point each on card data and residency |
The quotable version: Sierra vs Lorikeet for regulated industries scores 15 to 14 on the Eight-Gate Regulated-Fit Test; Sierra’s edge is card payments and hosting regions, plus FedRAMP High and AIUC-1, and Lorikeet’s is a public BAA commitment.
Two cautions. A score records what a vendor publishes, not what it will sign, so a 1 is a question, not a defect. And both vendors’ audit reports sit behind an NDA or request gate; we read the lists, not the reports.
Which vendor can take card payments inside the conversation?
Sierra can, and documents it in detail. Its trust and reliability page says “Sierra holds PCI DSS Level 1 Service Provider certification” and that payment data “flows through dedicated PCI certified infrastructure and never touches Sierra’s core platform, LLMs, or persistent storage.” Its 7 April 2026 payments announcement explains the mechanism: “the agent switches to a secure transaction flow which removes the agent while the processor completes the payment”, card details arrive by keypad on voice or an embedded form on chat, and the agent “only receives non-sensitive data like payment status and last four digits.” The Attestation of Compliance and a Shared Responsibility Matrix are available through its trust centre, whose update log records a PCI DSS v4.0.1 AOC on 1 November 2025.
Lorikeet takes the other route and says so. Its FAQ states that it “doesn’t store or process raw payment card information”, works with processors’ secure tokens and logs every payment-related action. Its own guide for lenders is more direct still: “Lorikeet does not hold PCI DSS certification today, so lenders that require the agent itself to be in PCI scope for card capture will need a compliant payment processor in the flow.” The same paragraph says Lorikeet publishes no uptime SLA.
Which matters depends on your payment flow. If customers will read or key card numbers to the agent, the vendor handling that data needs its own attestation in your vendor file, and Sierra has one. If your design keeps the card number out of the conversation, with a payment link or your processor’s hosted fields, the agent only sees tokens and may sit outside your cardholder data environment; your QSA decides that. Sierra’s attestation has a limit: its shared responsibility matrix exists because some controls stay with you. We set out the scoping routes in our guide to whether an AI voice agent can take card payments and what that does to PCI scope.
Which vendor publishes a HIPAA business associate agreement?
Lorikeet states it on three public pages. Its trust page says “We sign Business Associate Agreements for healthcare customers”, its healthcare page repeats it, and its FAQ says the AI only accesses the protected health information (PHI) each task needs, “following the minimum necessary standard”, with logging of PHI access. Its pricing table lists a DPA or BAA (including HIPAA) as standard on Start and Scale, custom on Signature.
Sierra lists HIPAA on its trust and reliability page and trust centre, which holds a “Sierra HIPAA Audit Report (2026)” and a paper titled “Enabling Healthcare: HIPAA for AI Agents”, both behind a request-access gate. Its trust and reliability page, healthcare page, trust centre overview, certification announcements and privacy policy make no public statement that Sierra signs business associate agreements; its documentation needs a sign-in, so that rung could not be walked. That is a gap in what is published, not evidence Sierra will not sign one: its healthcare page describes prior authorisation and claims work that, for a HIPAA covered entity, would normally involve PHI and therefore a BAA.
Why the BAA matters more than the word “HIPAA” on a trust page: under 45 CFR 164.502(e), a covered entity may disclose PHI to a business associate, and let it create, receive, maintain or transmit PHI on the entity’s behalf, if the covered entity obtains satisfactory assurance that the associate will appropriately safeguard the information. The covered entity does not have to obtain that assurance from a business associate that is a subcontractor; instead, the business associate obtains it from its own subcontractor. In both cases the assurance must be documented through a written contract or other written agreement or arrangement that meets 45 CFR 164.504(e). Our walkthrough of the HIPAA BAA chain for AI voice agents, and who signs what, shows how to map that chain before a pilot.
Which guardrails does each vendor use for regulated conversations?
Both pair rules a model cannot argue past with checks that need judgement.
Sierra’s version has four layers, set out in its 13 August 2026 defence-in-depth post: grounded content, natural-language policies, supervisors (“separate models that audit the conversation in real time, blocking, rewriting, or nudging the agent”), and deterministic guards, such as “preventing account-modification tools from even being available to the agent until the customer has been authenticated.” Its enterprise-grade agents post adds that output supervisors can switch “from observe to intercept” on high-risk topics. Sierra is candid about the limit: supervisors, “as agents themselves they share some of the limitations they’re built to detect”, and deterministic guards “only fire on what’s been explicitly defined.”
Lorikeet’s guardrails page opens with “Guardrails that assume the AI can be tricked”. Customer isolation, server-side identity validation, scoped tool access and execution caps are “enforced in code rather than prompts”; a prompt-injection classifier screens input; responses can be blocked, rewritten or escalated. It states prompt injection cannot be prevented entirely and aims for “the blast radius of any successful manipulation” to be “a single conversation.” Its how-it-works page says payments and identity checks “run as code the agent can invoke but never alter”.
One difference: Sierra’s guardrails are independently tested on a schedule. Its AIUC-1 certification (17 September 2026) followed adversarial testing and an audit by Schellman, and Sierra says the evaluations “recur at least quarterly, with a full audit every year”. Lorikeet tests its guardrails through simulations and guardrail test scenarios, and its trust page lists third-party penetration testing, but no third-party AI-agent certification such as AIUC-1 is listed on its trust page or trust centre as at 4 October 2026.
What audit trail will a regulator or examiner get from each?
Both vendors document audit evidence in detail and both score 2. Lorikeet’s FAQ lists “audit trail exports, compliance dashboards, exception reporting, customer consent reports, data access logs, incident reports, and custom reports” for regulatory reporting, and says every action keeps a complete audit trail and can require human authorisation in high-value scenarios. Its 11 March 2026 release note adds that any configuration change made through its MCP server “is logged with a full audit trail”, with role-based access control.
Sierra documents three pieces. Agent Traces (1 October 2025) record “the step-by-step decision path of every agent message in every conversation”, in production and testing. Release governance (20 August 2026) adds a Reviewer role, so a compliance stakeholder can be required to sign off a change after reading line-by-line diffs, plus immutable release snapshots for rollback. And its 22 September 2026 post “Your agent, laid bare” says “Conversation logs and performance data are available, and export APIs let you feed that information into your own data warehouse, BI tools, or reporting systems” and that the agent “lives in a Git repository you can access, giving you a history of what changed”. The narrower difference: Lorikeet names data access logs, consent reports and incident reports, and Sierra’s public pages list none of those three as a report a customer receives. Ask both for a sample export: a listed report type is not a demonstrated one.
The quotable version: both vendors let you export the record of what the agent did; Lorikeet also names data access logs and consent reports as customer reports, which Sierra’s public pages do not.
How does each vendor hand off to a human?
Both score full marks here. Sierra’s Agent SDK page says the agent “intelligently routes the conversation to the right team member, automatically generating a detailed summary for a smooth handoff”, its voice page promises “full context and history”, and Live Assist puts the agent’s knowledge in front of human reps.
Lorikeet’s FAQ lists the triggers: no matching workflow, missing information, a sensitive topic, or a customer asking for a human. Escalations route by issue type “with full audit trails showing exactly why each escalation occurred” and are not charged, and Lorikeet says it “doesn’t make clinical judgments”. Its 5 March 2026 Resolution Loop release lets a human take over in the same thread, with a “Steer” mode described as rolling out.
The real handoff test is whether the record of why escalation happened survives. Lorikeet names an audit trail for each escalation; Sierra traces every agent message and its conversation logs can be exported, so check the escalation reason is captured in that trace.
Can customer data stay in Australia or the EU?
Both vendors name Australian storage, and neither names Australian model inference for all providers. The detail is in the subprocessor lists (Sierra, Lorikeet), and it is the citable count on this page.
| Subprocessor list, read 4 October 2026 | Sierra (14 listed) | Lorikeet (20 listed) |
|---|---|---|
| Listed with an Australian location option | 2 (AWS; Cloudflare) | 0 by name; storage residency in Australia stated separately on its trust page |
| Listed with US and EU location options | 9 (Google, Microsoft, OpenAI, Deepgram, Baseten, Fireworks AI, Modal, Together, ElevenLabs) | 0 |
| Listed in the United States only | 3 (Anthropic, Twilio, WorkOS) | 16 (including OpenAI, Anthropic, Baseten, Groq, Deepgram, ElevenLabs, LiveKit, Twilio) |
| Listed as global, or with no location shown | 0 | 4 (AWS global; Google Cloud Platform, Cerebras and Braintrust with no location shown) |
Read it this way. Sierra’s AWS entry (“data storage, cloud hosting, and AI services”) has an Australian option, its other AI subprocessors US or EU, and its ISO announcement says customers “retain control over their data—where it lives, how it’s used, and when it is deleted.” Lorikeet’s trust page offers “data storage residency” in Australia and the EU from US primary hosting, its healthcare page says “AU/NZ data residency”, but its pricing table lists USA for both published plans, and its located inference providers are all in the USA. Ask both where prompts are processed, not only where transcripts are stored. Our guide to data residency for AI voice agents, hop by hop lists every hop to check.
Which is better for fintech, healthcare and insurance?
| If your non-negotiable is… | Shortlist first | Because (vendor’s own page) | Ask before signing |
|---|---|---|---|
| Card capture inside the conversation (cards, bill pay, collections) | Sierra | PCI DSS Level 1 Service Provider; AOC in trust centre | The shared responsibility matrix, and which controls stay with you |
| Card payments by processor token or payment link only | Either | Lorikeet documents the token pattern; Sierra’s payment flow also hands off to your processor | Your QSA’s view on whether the agent sits in scope |
| A signed BAA for US health data before the pilot | Lorikeet | Publicly states it signs BAAs; DPA or BAA standard on Start and Scale | Sierra: request the trust centre HIPAA paper and its BAA terms |
| Named data access logs, consent reports and incident reports | Lorikeet | Its FAQ lists all three for regulatory reporting; Sierra documents conversation-log export APIs but lists none of the three as customer reports | Both: a sample export from a test conversation |
| Independent AI-agent testing or an AI management standard | Sierra | AIUC-1 (testing at least quarterly) and ISO/IEC 42001:2023 | Scope of the AIUC-1 test against your own journeys |
| US federal or government work | Sierra | FedRAMP High, announced 10 June 2026 with Knox Systems | Which environment your agency’s data will run in |
| Insurance claims with brokers, assessors and claimants | Either; Lorikeet documents multi-party claims coordination | Lorikeet insurance page: specialist agents coordinate across assessors, insurers, claimants and brokers; Sierra lists quote and bind, policy servicing and FNOL | A recorded FNOL walkthrough on your own policy rules |
| Price you can model before a sales call | Lorikeet | Start $2,100 and Scale $5,100 a month, paid annually | Sierra: the agreed outcome definitions and any blended fee |
In fintech the deciding gate is usually card data, settled by Sierra’s attestation if card numbers pass through the agent. In US healthcare it is usually the BAA, where Lorikeet’s public commitment shortens the first conversation, though Sierra’s HIPAA audit report suggests similar paper exists on request. In insurance, neither lists an insurance-specific certification, so compare the audit record each leaves on the same claims scenario.
Which deployment model does each use, and who builds the agent?
Both are hosted platforms. Sierra holds FedRAMP High and offers three build routes: Agent Studio “without writing a line of code”, the Agent SDK to “develop agents in your existing programming environment and software development lifecycle”, and its agent development team, which it says has “done hundreds of deployments”. Lorikeet runs on Google Cloud with no public internet access to production and tenant isolation, plugs into your helpdesk, and adds dedicated deployment resourcing (people, not separate infrastructure) from the Scale plan upward.
Neither vendor’s public pages, as read on 4 October 2026, describe running the agent or its models inside your own infrastructure. Sierra’s “Your agent, laid bare” post says an agent’s logic can be exported “in a clearly structured format that’s portable to other platforms” and its code exported from Git, which is portability rather than self-hosting. If your regulator or board requires that, the question moves to a different category of platform, and the trade-offs are in our explainer on private AI deployment for AI sales agents.
Comparing AI support agents because your regulated business also sells by phone? Zian AI runs autonomous agents across phone, SMS, email and WhatsApp in 30+ languages, with a niched bank KYC onboarding agent and private model deployment on your own infrastructure. Zian AI is in partnership-application beta.
How to run this comparison yourself in one week
Your decision should rest on paper the vendors send you, not public pages. A one-week sequence:
- Day 1: request both trust centres’ reports, and from Sierra the PCI AOC and shared responsibility matrix.
- Day 2: map one regulated journey (a dispute, prior authorisation or FNOL) and mark where card data, PHI or identity data appears.
- Day 3: send both the same four written questions: BAA or DPA terms, where prompts are processed, what an audit export contains, and who can approve a change.
- Days 4 and 5: run the same scripted journey on both, including an attempt to act before authentication, and keep each trace or export.
- Days 6 and 7: re-score the eight gates from that paper. A 1 confirmed in writing becomes a 2; a 2 the paper does not support drops.
It costs most of a week of a security analyst’s and a contact-centre lead’s time, and it is the version your auditor will accept.
Frequently asked questions
Is Sierra or Lorikeet better for regulated industries?
Sierra edges it: on our eight-gate regulated-fit test it scores 15 of 16 and Lorikeet 14 as at 4 October 2026. Sierra scores higher on card data and residency, with a PCI DSS Level 1 service provider attestation and named regions including Australia, and also lists FedRAMP High and AIUC-1. Lorikeet scores higher on health data because it publicly states it signs BAAs, and it names data access logs and consent reports.
Does Sierra sign a HIPAA business associate agreement?
Sierra lists HIPAA and a 2026 HIPAA audit report in its trust centre, but the public pages we read on 4 October 2026 do not state that it signs BAAs, so ask for its terms. Lorikeet states publicly that it does. The agreement matters because 45 CFR 164.502(e) lets a covered entity share protected health information with a business associate if it obtains satisfactory assurance, documented in a written contract or other written agreement or arrangement, and a business associate that passes PHI to its own subcontractor must obtain the same assurance from that subcontractor.
Can Sierra or Lorikeet take card payments during a call or chat?
Sierra can: it holds PCI DSS Level 1 service provider certification, captures card details by keypad or embedded form, and passes only payment status and the last four digits back to the agent. Lorikeet says it does not store or process raw card data, works through your processor’s tokens, and states that it does not hold PCI DSS certification today.
Can I keep customer data in Australia with Sierra or Lorikeet?
Storage, yes, with both. Sierra’s subprocessor list offers Australian localisation for AWS and Cloudflare. Lorikeet’s trust page offers storage residency in Australia and the EU, though its pricing table lists USA for its two published plans. Neither lists Australia for all of its model providers, so ask where prompts are processed, not only where transcripts are stored.
Which is better for insurance claims, Sierra or Lorikeet?
Both name first notice of loss and policy servicing. Lorikeet’s insurance page describes specialist agents coordinating across assessors, insurers, claimants and brokers, while Sierra lists quote and bind, policy servicing and FNOL. Neither lists an insurance-specific certification, so run the same claims scenario through both and compare the audit record each leaves.
Do Sierra and Lorikeet publish their prices?
Lorikeet does: Start is $2,100 a month and Scale $5,100 a month, both paid annually in US dollars, plus a rate per resolution; Signature is custom with a $75K platform fee. Sierra publishes no prices and uses outcome-based pricing agreed per customer, with a blended model where outcomes do not fit.
How we read the vendor pages
Vendor facts come from sierra.ai, trust.sierra.ai, lorikeetcx.ai and trust.lorikeetcx.ai, read on 4 October 2026. The JavaScript trust centres were rendered in a headless browser. Before recording any absence we confirmed a known string on the same saved page. Sierra’s documentation and both vendors’ audit reports need a sign-in or request, so those rungs could not be walked.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Sierra 15 of 16, Lorikeet 14 of 16, Eight-Gate Regulated-Fit Test | Zian AI (our scoring of the vendor pages below) | This page, method in the scoring section | 4 October 2026 |
| PCI DSS Level 1 Service Provider | Sierra | https://sierra.ai/product/trust-and-reliability | 4 October 2026 |
| PCI DSS v4.0.1 AOC, published 1 November 2025; SOC 2 Type II and HIPAA reports updated 19 February 2026; 11 compliance items; 14 subprocessors and their locations | Sierra (Vanta trust centre) | https://trust.sierra.ai/ | 4 October 2026 |
| Payments launch 7 April 2026; agent receives payment status and last four digits | Sierra | https://sierra.ai/blog/payments | 4 October 2026 |
| FedRAMP High, announced 10 June 2026; hundreds of deployments | Sierra | https://sierra.ai/blog/certified-fedramp-high | 4 October 2026 |
| AIUC-1, announced 17 September 2026; evaluations at least quarterly, full audit every year | Sierra | https://sierra.ai/blog/sierra-achieves-aiuc-1-certification | 4 October 2026 |
| ISO 27001 and ISO 42001 announcement | Sierra | https://sierra.ai/blog/sierra-is-now-iso-42001-and-iso-27001-certified | 4 October 2026 |
| Release governance, 20 August 2026 | Sierra | https://sierra.ai/blog/release-governance-guardrails-for-agents-at-scale | 4 October 2026 |
| Agent Traces, 1 October 2025 | Sierra | https://sierra.ai/blog/agent-traces | 4 October 2026 |
| Your agent, laid bare, 22 September 2026: conversation-log export APIs; Git change history | Sierra | https://sierra.ai/blog/your-agent-laid-bare-and-why-it-matters | 4 October 2026 |
| Defence-in-depth post, 13 August 2026 | Sierra | https://sierra.ai/blog/defense-in-depth-in-the-age-of-agents | 4 October 2026 |
| Sydney office, March 2026 | Sierra | https://sierra.ai/blog/sierra-comes-to-australia | 4 October 2026 |
| SOC 2 Type II, ISO 27001:2022, HIPAA, GDPR; TLS 1.3 and AES-256 | Lorikeet | https://www.lorikeetcx.ai/product/trust | 4 October 2026 |
| 20 subprocessors and their locations; 2025 audit reports | Lorikeet (Vanta trust centre) | https://trust.lorikeetcx.ai/ | 4 October 2026 |
| $2,100 and $5,100 a month (paid annually, USD); $75K Signature platform fee; residency USA, USA, Custom | Lorikeet | https://www.lorikeetcx.ai/pricing | 4 October 2026 |
| PCI DSS not held; no uptime SLA published | Lorikeet | https://www.lorikeetcx.ai/articles/best-compliant-ai-collections-lenders-2026 | 4 October 2026 |
| Release notes: Lorikeet MCP 11 March 2026; Resolution Loop 5 March 2026 | Lorikeet | https://www.lorikeetcx.ai/release-notes | 4 October 2026 |
| 45 CFR 164.502(e) business associate assurances | US Government (eCFR) | https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-C/part-164/subpart-E/section-164.502 | 4 October 2026 |
Building a regulated phone, SMS or WhatsApp workflow that sells as well as serves? Zian AI’s agents handle live phone, SMS, email and WhatsApp outreach, integrate with HubSpot, Salesforce, HighLevel and Zapier, and can run on private model deployment on your own infrastructure. Zian AI is in partnership-application beta.