Clio Care

Clio Care is documentation software for independent NDIS support workers. After a shift, the worker describes what happened in their own words, and Clio's AI turns it into a structured progress note using only what they described. The worker reviews, edits and signs every note. Clio also uses AI to write reports from signed notes, suggest improvements to a worker's writing, read receipts for tax records, set up participant profiles from uploaded plan documents, and answer questions in its in-app help and website chat.

Pre-built models & their purpose

AI documentation for NDIS support workers. We use Anthropic's Claude models to draft case notes and reports, read receipts and uploaded plan documents, and answer help questions. We do not train our own models.

Other risk areas

AI-assisted documentation about people with disability (NDIS progress notes). We treat this as higher risk because the records can inform people's supports. Mitigations: the AI only prepares drafts and never makes decisions; every note is reviewed and signed by the worker; missing details are marked for the worker rather than invented, and block sign-off until completed; notes are tested against our published Note Integrity Standard before every release.

We keep a risk register covering fabricated content, missed incident flags, errors in plan goals, privacy, biased language, over-reliance by workers and defects in our own code, each with controls and monitoring. Every change to how Clio writes notes is assessed first: we write test scenarios for it, run our full test suite including anti-fabrication tests, check it on a staging version, and release only if every test passes. After release, we check real notes for known failure patterns using de-identified data. For example, in September 2026 a check found notes where a plan goal had been cut short. We traced it to a cleanup rule, fixed it, confirmed the fix with tests and released it.

Frequency & method

Yes. A person reviews every AI output: the worker reads, edits and signs every note before it is final. Every change to note generation passes our test suite before release, and we check real notes for known problems using de-identified data. The Founder reviews our AI risks, policies and providers at least once a year.

Example

Yes. Clio's AI does not make decisions; it prepares drafts. Workers can edit, regenerate or discard anything it produces, and a note is only final when the worker signs it. Anyone with a concern about a Clio-assisted note or how our AI behaves can email hello@cliocare.com.au, and we aim to respond within five business days.

Frameworks followed

Responsible for oversight

Yes. The Founder oversees all AI development. Every change is recorded in our development records and version control, tested with automated test suites, and reviewed on a staging version before release. Our governance framework, Responsible AI Policy, privacy policy and data processing summary are kept up to date.

Methods

Other methods

Participant records are stored in Australia, encrypted at rest (AES-256) and in transit (TLS), with each account limited to its own data. Our servers also run in Australia. When a task needs AI, the information for that task goes to our AI provider in the United States under a data processing agreement: it is deleted within 30 days and never used to train AI models. Workers are told this and asked to agree before first using AI features. We follow the Australian Privacy Principles, keep a data breach response plan, and use de-identified data for our own analysis.

Methods

Other methods

Clio does not infer or record personal characteristics about participants. The participant profile asks for the participant's pronouns, including she/her, he/him, they/them or any other pronouns the worker enters, and Clio always uses the pronouns recorded. Only when none have been recorded does Clio use pronouns suggested by the participant's name, so notes read naturally. Clio follows person-first language rules and replaces a list of deficit-based terms. Every note is reviewed and signed by the worker, who can correct anything before it is final.

Low-scale use. AI only runs when a worker asks for it, we do not train our own models, and we use established cloud providers rather than our own infrastructure. Our servers run in Sydney, next to our database, which avoids unnecessary data transfer.

Independent support workers spend hours of unpaid time on paperwork. Clio gives that time back while producing accurate, respectful records about the people they support, which participants rely on for their plans and supports. As we grow, we will add independent technical review of changes to our AI, continue reducing the information sent to our AI provider, and publish updates to our Note Integrity Standard as it evolves.