Why AI note-taking tools are not always safe for law firms to use 

“I like writing file notes and filling in timesheets,” said no solicitor, ever. 

Recording time and capturing what was said have always been necessary chores. Tools that automate this job by transcribing phone calls or meetings then saving these notes into management systems look like an easy, low-stakes way to bring AI into a practice.  

Using AI to draft court materials and contracts requires a high level of due diligence to catch errors and hallucinations.  While transcription feels safer, the ice under our feet may be thinner than it looks. This isn’t a low-risk decision.  

A transcription or telephony tool hears everything: the privileged advice, the confidential settlement figure, the client who should not have admitted what they just did. Nothing in a practice has broader reach into protected material than a service that listens in on every call. The same capacity that makes these tools convenient leads to significant exposure.

That makes the choice of provider an important professional decision. It does not matter how well the tool works if using it breaches your duties to your client and the court.

At stake: your client’s data

A solicitor must not disclose information confidential to a client and acquired during the engagement except as the conduct rules permit. The rules allow us to use third party providers to process and store data electronically. No modern firm can really survive without doing so. However, these providers must be selected carefully – especially in the wild west of the AI startup. 

Privacy law may also apply. Where a provider stores or processes regulated personal information outside Australia, the practice remains accountable under the Privacy Act (Section 16C) for how the overseas recipient handles it and has a duty to make disclosure and proper enquiries first.

Guidance Statement No 37 reminds practitioners that confidentiality obligations apply to AI tools, and that solicitors should take reasonable steps to ensure a tool does not misuse or expose confidential information. 

To make the selection task a little easier, QLS and the other Australian law societies released a joint AI selection and use checklist available here

Think like the lawyer you are

Not all providers meet the above requirements and only a careful (and rather cynical) analysis of the terms of the contract will reveal if they do. Several criteria must be examined with special care: 

1. What is actually done with the data? Is the raw information or transcription output used to train models, “improve the service”, or for any other purpose? A “no training” promise is a good start but needs close examination. Some wording is less credible than it appears: “We do not train on confidential data” may rest on the fiction that stripping names removes confidentiality.  “We do not train on your data” is of little use if “We may share data with trusted partners” and little is known about those partners or how they will handle your client’s information.
2. Who the data is shared with, and on what terms therefore becomes a critical issue. Professional know-how is valuable precisely because it is hard to obtain, and there is a real market for it. Many AI providers (quite sensibly) use third party AI models to power their system. This is not a problem in itself; it all comes down to the agreements between these entities.
3. Where is the data kept or processed? Storage in Australia avoids the cross-border accountability and disclosure problem arising from the Privacy Act; storage offshore engages it.
4. Ask about security in concrete terms. A recognised certification such as ISO/IEC 27001 or SOC 2 reflects independent assessment. A statement that the provider takes security “very seriously” or works towards a framework “aligned with” a standard, is not the same thing. “Hosted on AWS” is fine, but the hosting services are not responsible for security. They are only supplying a platform.
5. Read the contract, not just the marketing message. The website FAQ is not the agreement; the combined user agreements, terms of use and privacy policy are where you should look. Where there is divergence between the marketing and the contract, the contract terms govern. “I didn’t read it” is a poor defence for a lawyer.
6. Consider how willingly the provider answers questions. Transparent, specific responses are a good sign. Evasion can be an answer in itself.

Practice tips

• Treat the selection of a transcription, recording or telephony tool as a confidentiality and privacy assessment, not just a software purchase. 
• Use the joint AI selection and use checklist, and keep a record of the assessment. 
• Read the binding user agreement, not just the website, and resolve any inconsistency in favour of the contract. 
• Confirm data residency and whether the provider or any downstream processor(s) trains on, retains, shares or has other licences for your (and your client’s) data. 
• Require evidence of certification, not aspiration. 
• If the provider will not answer your questions willingly, consider that a reason to think about looking for an alternative. 

So, can my practice use transcription tools or not?

With care, yes. There is nothing wrong with using AI to take the drudgery out of file notes and time recording. The risk lies in assuming the task is low stakes because it feels administrative. AI is a new technology, and many of the providers involved do not have the same level of security and confidentiality that we are used to from legal software vendors. Just because the system can integrate with your trusted practice management system does not mean proper due diligence has been applied. That is your responsibility.

The data involved is among the most sensitive a practice holds, and the duties attached to it do not lessen because the tool is convenient. Do the due diligence before the first call is recorded, not after you are explaining yourself to a regulator.