Is pasting client data into AI a notifiable breach?

It can be. The test is whether personal information went somewhere it should not have and serious harm is likely. An accidental paste can satisfy both.

Short answer: It can be, and nothing about it being accidental changes that. Australia's Notifiable Data Breaches scheme turns on two questions: whether personal information was accessed or disclosed without authorisation, and whether serious harm to a person is likely. An unauthorised disclosure includes sending information somewhere it should not go. It does not require an attacker.

Key takeaways

First: does the scheme even apply to you?

Do not skip this. Telling a business it must notify when it has no such duty is as unhelpful as the reverse.

The Notifiable Data Breaches scheme sits inside the Privacy Act 1988 and applies to entities the Act covers. A business with annual turnover of $3 million or less is generally not covered, which describes a large share of Australian firms of three to twenty five people.

The exceptions matter and catch ordinary practices. Health service providers holding health records are covered regardless of turnover. So are businesses trading in personal information, contractors under a Commonwealth contract, credit reporting bodies, and reporting entities under the anti money laundering regime, which since 1 July 2026 includes accountants, conveyancers, lawyers and real estate professionals in relation to that work. You can also opt in voluntarily.

We work through coverage in detail separately. The short version: check before you assume, because the answer is yes more often than owners expect.

If you are in New Zealand, stop wondering. The Privacy Act 2020 binds every agency with no size threshold, and the notification duty applies to a sole trader.

What actually counts as a notifiable breach

Assuming you are covered, an eligible data breach has two elements.

Unauthorised access, unauthorised disclosure, or loss of personal information. Note the word disclosure. This is the element people misread. It does not mean somebody took it. It means the information went somewhere it was not authorised to go. Emailing a client list to the wrong address is a disclosure. So is pasting a client's identity documents into a service the client never agreed to and your engagement terms never contemplated.

Serious harm to one or more individuals is likely. Likely means more probable than not. Harm includes identity theft, financial loss, threats to physical safety, humiliation and damage to reputation or relationships.

Both elements have to be present. This is why the answer to the headline question is genuinely "it depends" rather than a dodge.

Applying it to a real paste

Consider four cases at the same firm.

A staff member pastes a paragraph of an anonymous case description. No personal information that identifies anyone. Not a breach.

A staff member pastes a client's name and the fact that they are a client. Personal information, disclosed. Serious harm unlikely in most contexts. Assess and record your reasoning, but this probably does not qualify.

A bookkeeper pastes a bank statement showing account numbers and transaction history. Personal and financial information. Disclosure to a third party service outside your engagement. Likely serious harm is arguable and in many cases yes. This one needs a proper assessment.

Someone pastes a client's passport and licence details. Identity documents are the classic serious harm case, because the person cannot change them and they are the raw material for identity theft. Treat as likely notifiable and start assessing immediately.

The pattern: the more the exposed data enables impersonation of the person, the closer you are to notifying.

The clock, and what to do first

If you suspect an eligible breach, you must carry out a reasonable and expeditious assessment, and complete it within 30 days. If you conclude it is an eligible data breach you must promptly notify the Commissioner and the affected individuals.

In New Zealand, if a privacy breach has caused or is likely to cause serious harm, you must notify the Privacy Commissioner and affected people as soon as you are practically able.

Before any of that, do the containment. For a paste into an AI tool that means: delete the conversation and any uploaded file where the product allows it, check whether the account's history and training settings retained it, record exactly what was sent and when, and if any credential was included, rotate it now rather than after the assessment. A rotated key stops mattering. A disclosed passport number does not.

What the numbers say about how this happens

The OAIC received 1,205 data breach notifications in the 2025 calendar year, up 8 per cent on 1,112 in 2024 and the highest annual figure since the scheme commenced in 2018. Most were attributed to malicious or criminal activity, with 716 notifications.

That leaves a substantial remainder that were not attacks. Human error has been a persistent and material share of notifications every period since the scheme began. Nobody breaks in; someone sends something to the wrong place. An accidental paste belongs to that family, and it is the family that is easiest to reduce.

Reducing the odds without banning the tools

Banning AI does not work. People use their phones instead and you lose the visibility you had. The workable approach is to name the categories that must never go in, provide business accounts so there is one set of terms to reason about, and make the risky moment visible when it happens.

Prinn does the last part. It runs on each person's computer and checks a prompt before it is sent. If it finds an identity document number, a bank account, a card number or an API key, it shows what it found and offers three choices: redact it, stop, or send it as is. The person decides, and it never blocks anyone. Prompt content never leaves the machine, including to us.

The part that matters for this page is the record. An owner sees the category, the tool, the date, the device and the choice made, without reading anyone's prompts. If you ever have to assess a suspected breach against a 30 day clock, knowing what was actually sent and when is the difference between an assessment and a guess.

It is reliable on credentials, financial details and identity numbers. Names, home addresses and dates of birth written into ordinary prose are harder, and it is still improving at those. Prinn is not legal advice, does not make you compliant with anything, and cannot tell you whether a given incident is notifiable.

Related

Sources