Claude data residency in Australia: onshore

Your board asks where the AI keeps its data, and for once there is a clean answer available. Claude data residency in Australia is real, specific and buyable today, but only on one of the three paths into the model. The consumer and business apps process conversations on Anthropic's infrastructure overseas, while Amazon Bedrock runs Claude inference, the actual processing of your prompts, inside Sydney data centres in the ap-southeast-2 region. Australians have adopted the tool at scale, with Australia ranking fourth globally for Claude.ai usage relative to population, and Anthropic opened its Sydney office in March 2026¹. The infrastructure followed the users. This guide maps what stays onshore on each path, what the Privacy Act actually asks of you, and when paying for Sydney inference earns its keep.

---

What does Claude data residency in Australia actually cover?

Data residency means the place where your prompts and outputs are processed and stored, which for AI tools is rarely the country you sit in. Claude reaches Australian businesses along three paths, and each one keeps a different amount onshore. The path decides your Privacy Act homework.

Claude apps: claude.ai, Teams and Enterprise

The apps most people use, claude.ai in the browser plus the Claude Teams and Claude Enterprise plans, process conversations on Anthropic's own infrastructure, and Anthropic states that as a global company it may process data in the countries where it or its partners operate². No Australian region exists for the apps in 2026. Commercial plans still carry real protection, because Anthropic's commercial terms rule out training on your business data, a line that holds across Teams, Enterprise and the API³. For a marketing team drafting copy or an operations team summarising internal documents, that contractual layer is usually the whole story.

Anthropic API: the direct line

Developers calling the Anthropic API directly get the same overseas processing picture, with inputs and outputs deleted from Anthropic's backends within 30 days unless a zero data retention agreement, a contract under which prompts are never stored after processing, tightens it further³. Anthropic does not offer an Australian region for the direct API today. Latency from Australia stays workable for most workloads, and the 30-day deletion window satisfies plenty of legal teams. Regulated data is a different conversation.

Claude on AWS Bedrock Sydney: the onshore path

Amazon Bedrock, the AWS service that hosts foundation models inside AWS regions, has served Claude from Sydney since the upgraded Claude 3.5 Sonnet arrived in ap-southeast-2 in November 2024⁴. Requests sent to the Sydney endpoint are processed in Australian data centres, and newer Claude models add Australia geo inference profiles, routing rules that keep requests inside the country, which you confirm per model in the AWS regional availability tables⁵. Bedrock also sits in scope of the AWS IRAP assessment, the Australian Signals Directorate's security review programme, at the PROTECTED level alongside more than 150 other services, with the 2026 report downloadable from AWS Artifact⁶⁷. The assessment belongs to AWS, not to Anthropic. Pin the Australia or Sydney profile explicitly, because a global cross-region profile is allowed to route your request to any commercial AWS region on its list⁵.

---

How does APP 8 treat Claude under the Privacy Act?

APP 8 is the Australian Privacy Principle governing cross-border disclosure of personal information, and it opens with a distinction many teams miss. Routing data through an overseas provider counts as use rather than disclosure when your business keeps effective control through a binding contract, which is precisely the shape of a Claude Teams agreement or an AWS agreement⁸. Disclosure triggers the heavier duty: reasonable steps to ensure the recipient handles the information to APP standard, plus accountability under section 16C if the recipient breaches it anyway⁸.

Consent is the exception people reach for first, and it is narrower than it looks. An individual must be expressly told that APP 8 protection will not apply before agreeing, which is a hard sentence to put in a customer-facing form⁸. Contracts beat consent here. Keeping inference in Sydney shrinks the entire question, because personal information that never leaves Australia never engages APP 8 at all.

One date deserves a circle. From 10 December 2026, privacy policies must disclose automated decision-making that significantly affects individuals, so a Claude rollout that touches decisions about people needs its paperwork ready this year. We cover the wider legal picture in our guide to Claude and the Australian Privacy Act.

---

Practical steps before you pick a path

Checklist: an APP 8 answer you can defend

• Classify the data: split what Claude will touch into personal information, commercial-sensitive material and everything else, and write the three lists down.
• Match the path: everyday business content fits Claude Teams under the commercial terms, while health, financial hardship or government-classified material points to Bedrock Sydney.
• Paper the relationship: keep the Anthropic or AWS agreement on file as your effective-control evidence, since the use-versus-disclosure call rests on it.
• Pin the region: on Bedrock, select ap-southeast-2 or an Australia geo profile in code, and block global profiles in policy if residency is a requirement.
• Update the privacy policy: state where processing happens and prepare the automated decision-making disclosure before 10 December 2026.

Tip from satori.: classification is step one of our five-step AI approach for a reason. Teams that sort their data before choosing a platform make the residency decision in an afternoon, while teams that start with the tool argue about it for a quarter.

---

What breaks when you run Claude onshore

Honesty about the Sydney path saves budget arguments later. Bedrock gives you the model, not the product, so Projects, Claude Cowork and the polished apps stay behind, and someone builds or buys an interface before colleagues can type a prompt. Model availability in ap-southeast-2 can trail the US regions, which means the newest Claude release sometimes arrives in Sydney weeks after launch, and you plan upgrades around the AWS regional tables rather than Anthropic's announcement day⁵. Billing runs per token in US dollars, so finance needs a forecast rather than a flat licence fee.

Our own work runs on Claude Teams, because nothing we draft for clients requires Sydney inference, and that is the honest norm for consultancies and most professional services firms. The rule of thumb we give buyers is short. If your data classification says onshore, run Claude through Amazon Bedrock in ap-southeast-2 with an Australia geo profile pinned, and if it does not, buy Claude Teams at US$20 per user per month and spend the difference on training. Both paths can sit behind one managed Claude setup, where we handle licences, configuration and the Privacy Act paperwork for A$850 setup plus A$15 per user per month.

---

The bottom line

Claude data residency in Australia comes down to one architectural choice, made once and documented well. The apps and the direct API process data overseas under commercial terms that rule out training on your content, which satisfies the Privacy Act for most business use when the contracts are in order. Bedrock in Sydney processes prompts onshore, inherits AWS's PROTECTED-level IRAP scope and suits regulated data, at the cost of building your own interface and forecasting token spend. Satori configures both paths for Australian businesses at fixed prices, and you can book a free consultation to work through which one your data actually needs. Availability itself is not the hurdle, Claude is available in Australia in full, the residency choice is the decision that matters.

Open your privacy policy and count the sentences that describe overseas disclosure of personal information. Twenty minutes with that document tells you whether your current wording already covers an AI assistant, and the gap you find becomes the first line of your APP 8 assessment.

---

Sources

Footnotes

1. Anthropic (2026). Sydney will become Anthropic's fourth office in Asia-Pacific. https://www.anthropic.com/news/sydney-fourth-office-asia-pacific ↩

2. Anthropic Privacy Center (2026). How does Anthropic protect the personal data of Claude users?. https://privacy.claude.com/en/articles/10458704-how-does-anthropic-protect-the-personal-data-of-claude-users ↩

3. Anthropic (2026). API and data retention. https://platform.claude.com/docs/en/manage-claude/api-and-data-retention ↩ ↩²

4. Amazon Australia (2024). The upgraded Claude 3.5 Sonnet, Anthropic's most intelligent AI model to date, now available in AWS Sydney Region. https://www.aboutamazon.com.au/news/aws/the-upgraded-claude-3-5-sonnet-anthropics-most-intelligent-ai-model-to-date-now-available-in-aws-sydney-region ↩

5. AWS (2026). Supported Regions and models for inference profiles, Amazon Bedrock User Guide. https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html ↩ ↩² ↩³

6. AWS (2026). IRAP Compliance. https://aws.amazon.com/compliance/irap/ ↩

7. AWS Security Blog (2026). 2026 Phase 1a IRAP report is now available on AWS Artifact for Australian customers. https://aws.amazon.com/blogs/security/2026-phase-1a-irap-report-is-now-available-on-aws-artifact-for-australian-customers/ ↩

8. OAIC (2025). APP Guidelines Chapter 8: Cross-border disclosure of personal information. https://www.oaic.gov.au/privacy/australian-privacy-principles/australian-privacy-principles-guidelines/chapter-8-app-8-cross-border-disclosure-of-personal-information ↩ ↩² ↩³