At a glance: the short answer
A single AI phone call is not one system. It is six to eight separate services, each with its own region, and most residency promises cover only some of them. Audio, transcription, model inference, storage, CRM sync and error logging can each land in a different country. Ask for a named region per hop, in writing, rather than accepting one badge on a website.
The one-switch myth
Most vendor conversations about data residency end the moment someone says we host in Australia or we have an EU region. That is usually true and usually incomplete: it describes where a database sits, not where a call is reasoned about.
The gap is scope, not dishonesty. A voice agent vendor owns the orchestration layer and rents the rest: carrier minutes, speech-to-text, the language model, the voice, the error tracker. Each supplier has its own region map, defaults and definition of residency, so the promise is only as strong as the weakest link. Storage and processing are also different questions, and most buyers ask only the first.
Our voice AI vendor security questionnaire gives you twelve questions to put to a supplier, and the enterprise readiness checklist for AI sales agents covers residency alongside SSO, RBAC and audit logging. Our guide to AI data sovereignty in Australia sets out the Australian legal picture, including the coming automated decision-making transparency rules. This post does what none of the three does: it walks one call, hop by hop, naming who holds the data at each step and quoting what each supplier’s own documentation says about storage versus processing.
Walk the call: eight hops, eight possible jurisdictions
1. The telephony carrier and SIP trunk
Before your vendor’s software sees anything, a carrier has handled call setup and is bridging the media. Twilio’s global infrastructure documentation defines its Regions as “the data centers where your application’s Twilio data is processed and stored” and lists three: United States (US1), Ireland (IE1) and Australia (AU1). Two caveats sit on the same page: “During this initial phase of the Twilio Regions rollout, Twilio does not guarantee that all data will remain within your selected Region”, and “Not all products and features are available in Regions outside of US1.” A selector is not a guarantee.
2. Audio in transit and the recording buffer
Encryption in transit is table stakes. The residency question is where audio comes to rest, even briefly. Recording buffers, retry queues and media relays are all storage, and frequently sit in the vendor’s home region rather than yours. Ask whether recording is on by default and where the buffer lives.
3. Speech-to-text
Transcription turns raw voice into searchable text, usually the highest-risk artefact in the pipeline, and is commonly a different supplier from the model doing the reasoning, with a different region list. A vendor that has pinned its LLM to your region may still send audio to a transcription endpoint elsewhere.
4. The language model
This is the hop where in-region most often means something narrower than buyers assume, and the providers say so themselves.
Microsoft’s Foundry Models deployment-type documentation states that “Data stored at rest remains in the designated Azure geography”, then splits inferencing: Global types “May be processed in any Azure region”; Data Zone types process data “only within the Microsoft-specified data zone (US, EU, or Asia Pacific (APAC))”; Standard single-region types, “in the deployment region”. The same page tells you to “start with Global Standard” for most workloads and lists Global Standard as the recommended tier for the default case, so the path of least resistance is the one whose inferencing may be processed in any Azure region.
OpenAI’s data controls documentation repays close reading if you are Australian. Its residency table lists Australia (au.api.openai.com) as supporting storage but not regional processing, whilst Europe (EEA + Switzerland) supports both, and states plainly that “Support for regional storage does not imply support for regional processing.” The page states that “If your selected Region does not support regional processing, as identified below, OpenAI may also process and temporarily store Customer Content outside of the Region to deliver the services”, and that “Data residency does not apply to system data, which may be processed and stored outside the selected region.”
So an Australian buyer can have OpenAI data residency switched on and still have inference happen offshore. Nobody is hiding this. It is simply not what data residency sounds like it means.
5. Text-to-speech
The reply must be spoken, so generated text and often a voice model go to a synthesis provider. ElevenLabs, a widely used provider, offers residency as an enterprise feature with locations in the EU, India and Singapore, the United States being the standard location and Australia not among them. Its data residency documentation says that “While storage will take place in the selected location, processing may nevertheless occur outside of the selected location, including by ElevenLabs’ international affiliates and subprocessors, for support purposes, and for content moderation purposes.” A residency selection may govern storage whilst processing sits elsewhere.
6. Transcript, summary and structured outcomes
The hop vendors answer most happily, because they own it. Get the region named, retention stated in days, and deletion on termination in writing.
7. CRM, calendar and downstream analytics
Once an outcome is written to a CRM, residency becomes that CRM’s problem and your contract with it. A summary pushed into a US-hosted CRM instance has left your region no matter how carefully the first six hops were configured.
8. Logs, traces and error monitoring
The forgotten hop, and often the leakiest. Application monitoring captures request and response payloads for debugging, which for a voice agent means transcript fragments, phone numbers and names. Sentry, a common error monitoring platform, documents two storage locations, the US and the EU, and states that “once selected, your data storage location can’t be changed. The only way to switch it is by creating a new organization.” So a vendor that set up monitoring years ago in the US cannot move it for you. Sentry also lists data that “may be stored in the US, regardless of your selected Data Storage Location”, including user accounts and integration metadata. There is no Australian option in that documentation.
What to ask at each hop
| Hop | Who actually holds the data | Typical default | What to ask | What a good answer sounds like |
|---|---|---|---|---|
| 1. Carrier / SIP trunk | Telephony provider, not your vendor | US region | Which region is our account in, and where is media bridged? | A named region code, plus which features are unavailable there |
| 2. Audio in transit, recording buffer | Vendor or carrier | Recording on, vendor home region | Is recording on by default, and where is the buffer? | Opt-in recording, buffer region named, retention in days |
| 3. Speech-to-text | A separate STT supplier | Provider default endpoint | Which supplier, which endpoint, and is audio retained? | Supplier named, regional endpoint named, zero-retention confirmed |
| 4. Language model | Model provider | Global routing | Is inference regional, or only storage? Show the deployment type. | A named deployment type and SKU, not a bare our data is in Australia |
| 5. Text-to-speech | Voice synthesis supplier | Storage in region, processing global | Does your residency cover processing or only storage? | An explicit split, and the configuration that limits processing |
| 6. Transcript and outcome storage | The vendor | Vendor primary region | Region, retention period, deletion on termination? | All three in the contract, not the sales deck |
| 7. CRM, calendar, analytics | You and your CRM vendor | Your existing instance | What fields are written, and where does that instance live? | A field-level list you can review and restrict |
| 8. Logs, traces, error monitoring | An observability vendor | US region, PII often captured | Which tool, which region, and is payload scrubbing on? | Tool named, region named, scrubbing rules shown |
The Australian legal layer: APP 8
This section is general information, not legal advice; your obligations depend on your circumstances.
Australian Privacy Principle 8 governs cross-border disclosure. The OAIC’s guidelines state at paragraph 8.2 that “before an APP entity discloses personal information about an individual to an overseas recipient, the entity must take such steps as are reasonable in the circumstances to ensure that the recipient does not breach the APPs in relation to that information”, and that “Where an entity discloses personal information to an overseas recipient, it is accountable for an act or practice of the overseas recipient that would breach the APPs (s 16C).”
Accountability staying with the discloser is the operative point: choosing a vendor does not hand the obligation to that vendor.
One nuance cuts both ways. Paragraph 8.14 notes that “in limited circumstances providing personal information to an overseas contractor to perform services on behalf of the APP entity may be a use, rather than a disclosure”, where a binding contract limits the provider to storing and ensuring access and “gives the entity effective control of how the personal information is handled by the overseas recipient”. That carve-out is written around storage. A provider running inference on your call content does more than store it, so do not assume it stretches to hop four.
The GDPR layer, and why in-region storage is not the test
GDPR Chapter V covers transfers to third countries. Article 44 provides that “Any transfer of personal data which are undergoing processing or are intended for processing after transfer to a third country or to an international organisation shall take place only if, subject to the other provisions of this Regulation, the conditions laid down in this Chapter are complied with by the controller and processor, including for onward transfers of personal data from the third country or an international organisation to another third country or to another international organisation.”
What residency marketing elides is what counts as a transfer. The European Data Protection Board’s Guidelines 05/2021, version 2.0 adopted 14 February 2023, set out three cumulative criteria and then state that “remote access from a third country (even if it takes place only by means of displaying personal data on a screen, for example in support situations, troubleshooting or for administration purposes) and/or storage in a cloud situated outside the EEA offered by a service provider, is also considered to be a transfer, provided that the three criteria outlined in paragraph 9 above are met.”
Example 11 makes it concrete: a third-country processor with no EU establishment remotely accessing EU-stored data for support results in a transfer under Chapter V. Data can sit on a disk in Frankfurt and still be transferred, because a support engineer elsewhere can open it. In-region storage is necessary, not sufficient; the question is who can reach it. Which is why hop eight matters so much.
The strongest form of residency control
Every control above is a promise about someone else’s infrastructure. The structural alternative is to run the model inside a boundary you control.
Zian AI supports private model deployment on customer infrastructure, keeping inference and conversation data inside your own environment rather than depending on a supplier’s region configuration. That addresses hop four, the hop with the least buyer visibility. It does not fix the other seven: your carrier, CRM and logging stack still need answering.
On the rest we would rather point at our own published wording. Zian’s security and privacy whitepaper states that “Machines are also hosted only in top-tier countries inside secure tier-1 server facilities in USA, Australia, and Europe”, and that SOC 2 Type II, ISO/IEC 27001, ISO 27017 and ISO 27018 are held by the underlying infrastructure and AI providers, so “we do not claim these certifications in Zian AI’s own name.” Read that as a description of where infrastructure sits, not as a region-selection feature or a residency commitment: neither is on offer in the beta. Demand the same precision from every vendor on your shortlist.
Zian AI has been running outbound acquisition since 2017 and is in waitlist beta: no public pricing, no free trial, no self-serve signup. For the broader Australian picture, see our guide to AI data sovereignty in Australia.
How we sourced this
Every external claim here was checked by opening the primary source on 28 August 2026: the OAIC’s APP guidelines chapter 8 for APP 8 and s 16C; EUR-Lex for the consolidated GDPR text of Article 44 (reachable and quoted directly, so no substitute was needed); the EDPB’s Guidelines 05/2021 version 2.0 PDF for the transfer criteria and Example 11; and the current public documentation of Twilio, Microsoft, OpenAI, ElevenLabs and Sentry.
What we did not verify: we have audited no vendor’s actual behaviour, only its published documentation, and documentation changes. The typical default column reflects what those pages describe at the time of writing; treat it as a starting point, not a substitute for asking. We have deliberately included no pricing figures, for Zian or any supplier named.
Frequently asked questions
Does storing our call data in Australia mean it never leaves the country?
No. Storage location and processing location are separate settings. OpenAI’s data controls documentation lists its Australia endpoint as supporting storage but not regional processing, and says that where a region lacks regional processing it “may also process and temporarily store Customer Content outside of the Region to deliver the services.” Ask both questions, separately, for every hop.
Which hop most often breaks a residency promise?
Two: model inference and error monitoring. Inference, because regional processing is something you have to opt into, and on some platforms be approved for; global routing is what you get otherwise. Monitoring, because it is configured once by engineers, usually in a US region, and never revisited at compliance review. Sentry’s documentation notes a storage region “can’t be changed” afterwards, so that one is hard to fix later.
Our call data is stored in Australia but processed overseas. Does APP 8 still apply?
Almost certainly. APP 8 turns on disclosure to an overseas recipient, not on where the database sits. The OAIC’s APP guidelines chapter 8 say that where an entity engages an overseas contractor to perform services on its behalf, “in most circumstances” the provision of personal information to that contractor is a disclosure, with a narrow carve-out at paragraph 8.14 for arrangements limited to storage where the contract “gives the entity effective control of how the personal information is handled by the overseas recipient”. Do not assume that carve-out covers a provider running inference on your call content. Either way, s 16C keeps the discloser accountable.
Under GDPR, is data stored in the EU but accessible from outside still a transfer?
Generally yes. The EDPB’s Guidelines 05/2021 state that remote access from a third country, including for support, troubleshooting or administration, is considered a transfer where the three cumulative criteria are met. Physical storage location alone does not settle the question.
Does private model deployment solve residency on its own?
It addresses the inference hop, the one you otherwise have least visibility into. It does not address your carrier, transcription, voice synthesis, CRM or logging stack. Treat it as the strongest single control available, not a complete answer.
Does Zian AI hold SOC 2 or ISO 27001 certification?
No, and we say so on our own security page. Those certifications are held by the underlying infrastructure and AI providers; Zian AI inherits the controls but does not claim the certifications in its own name. If a comparable vendor claims otherwise, ask to see the report and check whose name is on the cover.
Getting the answers before you pilot
Residency is not a badge, it is eight answers. If you are evaluating AI voice agents and need those answers mapped against your own obligations, that conversation is better had before a pilot than after one. Zian AI is in waitlist beta. Apply For Partnership and tell us which hops are gating for you.