Security and data

How we handle your data.

The agents run on your accounts, with the access you give them, and nothing goes out without a person approving it. This page sets out how, for you and for whoever looks after your IT, security or compliance.

Every fact on this page was last checked on 13 October 2026.

In short

Nothing goes to a customer, or changes in your systems, without a person on your team approving it.
Your content is never used to train AI models. We only build on Claude business plans, never personal accounts.
Everything runs on accounts you own. You can switch any agent off at any time.
Each agent reaches only what its job needs, and only what the signed-in person can already see.
Every release is tested on your own examples before your team sees it, and we show you the result.
We run Onwise on the same Claude business plan setup we build for you.

Control

A person stays in charge of everything the agents do. They prepare the work, your team decides what happens to it, and each agent is tested on your own examples before anyone relies on it.

Drafts, never sends.
Every email, CRM update and proposal waits for a person to approve it. Nothing goes to a customer, and nothing changes in your systems, until someone on your team says yes. We do not build anything that sends on your behalf.
Written limits for every agent.
Each agent comes with a short list of things it never does, such as making a commitment, quoting a price or promising a date on someone's behalf. The list is written down with you and travels with the agent.
Tested before your team sees it.
Before we build, we agree a set of real examples from your business and the answers you expect. Every release and every update is run against that set before it reaches your team, and we show you the result. The set grows as your process and data change.
Read-only where it should be.
On Claude business plans, an Owner can limit what a connected tool is allowed to do across the whole organisation, for example read the CRM but not write to it. People in your team cannot override that setting.

Data and security

The agents run inside your own Claude business workspace and your own tools, under Anthropic's commercial terms. Your content is not used to train models, each person only reaches what they could already reach, and you can see exactly where your data is stored.

Never used to train AI.
Anthropic's Commercial Terms say it may not train models on customer content from its business plans. The one exception is feedback a person chooses to send, such as a thumbs up or down, which Anthropic keeps for up to five years. Workspace Owners can switch feedback off. Personal Free, Pro and Max plans work differently, which is why we only build on business plans.
Runs on your accounts.
The agents work in your own CRM, inbox and AI workspace, with the CRM you already have. You own the accounts and the data in them, and you can switch any agent off at any time.
Only the access it needs.
You decide what each agent can see and learn, role by role. An agent gets the systems its job needs and nothing more. When a tool is connected, each person signs in to it themselves, and Claude only sees what that person is already allowed to see in that tool.
Clear about where data is stored.
Anthropic stores data from its business plans in the United States. Processing can run in the US, Europe, Asia or Australia. If your data has to stay in Australia, raise it on the first call and we will tell you plainly what that means for the build.
Clear about how long it is kept.
On Team and Enterprise plans, conversations stay available so people can pick them up again. A deleted conversation leaves the history straight away and Anthropic's storage within 30 days. Content flagged for breaking Anthropic's usage policy can be kept for up to two years. Enterprise plans can set their own retention period, with a minimum of 30 days.
Independently certified.
Anthropic holds SOC 2 Type I and Type II reports and ISO 27001:2022 and ISO/IEC 42001:2023 certifications, the last of which covers how AI systems are managed. Data is encrypted in transit (TLS 1.2 or higher) and at rest (AES-256).
Admin controls that come with the plan.
Team plans include single sign-on, domain verification and admin control over which tools can be connected. Enterprise plans add audit logs, automatic user provisioning (SCIM), role-based access, custom retention and a Compliance API for your own monitoring.

Compliance

We build to Australian privacy law and to the government's guidance on AI. That means we document what each agent uses and decides so your privacy policy stays accurate, we treat recording as something people agree to, and we follow the six practices the National AI Centre sets out for adopting AI.

Ready for the Privacy Act changes.
From 10 December 2026, your privacy policy has to say when a computer program uses personal information to make, or do something substantially related to making, a decision that could significantly affect a person. The policy has to name the kinds of personal information used and the kinds of decisions involved. The OAIC's guidance, updated on 30 September 2026, says this still applies when a person reviews or approves the output. Its examples include ranking job applicants and scoring loan risk. Most sales work, such as drafting emails, updating your CRM or preparing for a meeting, is unlikely to meet that test. Where an agent could come close to that line, we document the personal information it uses and the decisions it helps with, so your policy stays accurate.
Clear about data that leaves Australia.
Because Claude business plans store data in the United States, your business stays accountable under the Privacy Act for how that information is handled overseas. The OAIC asks businesses using AI products to check where the servers are and how the provider secures them, and to say in their privacy policy if personal information is likely to go overseas. We give you the facts on this page to do that.
Kept off personal AI accounts.
The OAIC recommends that businesses do not enter personal information, and particularly sensitive information, into publicly available AI tools, and that they check whether a provider can use what is entered to train its models. Our agents run on Claude business plans, where Anthropic's commercial terms bar training on your content and your organisation controls the workspace, not on personal accounts your team signed up for themselves.
Recording consent built in.
The rules depend on where each person is, and meeting capture tells people and asks first, wherever they are.
Everyone's consentNew South Wales, Western Australia, South Australia, Tasmania and the ACT generally need the consent of every party to record a private conversation, with narrow exceptions.
One party can recordVictoria, Queensland and the Northern Territory let a person who is part of the conversation record it, but limit what they can do with the recording afterwards.
Built to the government's AI guidance.
The National AI Centre's Guidance for AI Adoption, published 21 October 2025, replaced the earlier Voluntary AI Safety Standard. In December 2025 the National AI Plan set aside mandatory AI rules in favour of existing law, so this guidance is the reference for how to adopt AI responsibly. We follow its six practices.
  1. 01
    Decide who is accountable.Name the people who own each AI system and its outcomes. Each agent has an owner on your team as well as on ours.
  2. 02
    Understand impacts and plan accordingly.Work out who the AI affects and what could go wrong. The blueprint maps your process and the people in it before anything is built.
  3. 03
    Measure and manage risks.Find the risks that come with AI and put controls on them. Written limits and access kept to what each job needs are those controls.
  4. 04
    Share essential information.Be open about where and how AI is used. This page, and the notes we give you for your privacy policy, cover it.
  5. 05
    Test and monitor.Check the AI works before it goes live and keep checking after. Every release runs against your own examples first.
  6. 06
    Maintain human control.Keep a person able to oversee and step in. Every agent drafts and a person approves.

Bring your IT team's questions.

30 minutes on how the agents would run in your business, and what they would have access to.

See what we'd automate first