OPINION OF THE WEEK: Beware Of AI Notetaking Threat To Confidentiality - Wealth Briefing

Lawyers can put confidentiality and privilege at risk if AI notetakers are used in meetings – one of the risks of this burgeoning technology that HNW individuals need to understand.

In thinking about all the use cases and risks that attend AI in

today’s private client advisory sector, one term that ought to

stick in the mind right now is “confidentiality.”

For those who don’t follow the twists and turns of AI closely, as

well as legal cases, it might not appear obvious that one of the

dangers of AI if not used properly is the destruction of

lawyer-client privilege. In an industry where HNW and UHNW

individuals and families understandably value their privacy, this

is a big problem. 

Many of us who go on Zoom, Teams and other platforms for business

meetings might be told in advance – as we should – that “this

meeting is being recorded” – usually in a cheerful voice. For

journalists who haven’t mastered the grinding skill of shorthand

– as I did decades ago – all these recording tools are a

godsend. 

But there is a downside.

As noted by the law firm White & Case in an April 23 report,

entitled Attorney-client privilege and work product in the

age of generative AI, two cases illustrate the risks –

with very different results.  

On 10 February this year, the US District Court for the Eastern

District of Michigan in Warner v. Gilbarco,

Inc denied a motion to compel production of documents

that a self-represented litigant had prepared using a public AI

chatbot. The court found that work product protection

applied. 

A week later, on 17 February, the US District Court for the

Southern District of New York in United States vs.

Heppner reached the opposite result, ordering a criminal

defendant to produce documents he had generated using another

public AI chatbot while seeking legal advice. 

The Heppner case was also mentioned at a recent media webinar,

which I attended. The event was hosted by law firm ArentFox Schiff.

Sarah Severson, a partner, explained the risks of AI in meetings

where there are not clear understandings about what is

involved.

Notes taken via AI could be “discoverable,” she continued.

Automated transcripts and summaries are classed as digital

documents that opposing counsel can subpoena in civil or criminal

cases. 

Severson said lawyers should always disclose their use of AI to

clients in their engagement letters. 

“AI does not owe clients a duty of confidentiality or

accountability,” she said. 

The root of the problem is that cloud-based processing,

third-party vendor data access, and terms of service permitting

model training eliminate the legal expectation of

confidentiality. 

This is a global issue.

The UK is an example, as recent cases demonstrate. In the case of

UK v Secretary of State for the Home Department [2026] UKUT

81 (Hamid), the Upper Tribunal (Immigration and Asylum

Chamber) delivered the first decision by an English court or

tribunal to directly address what the legal professional

privilege risks when confidential and privileged material is

uploaded to open-source AI tools. 

It is important to remember that jurisdictions such as

Singapore share the common law traditions of the UK and US, for

example. As case law builds up around the world, the AI

notetaking issue will become a global one. 

One of the points coming out of all this is that – as Severson

said in the webinar – AI is a valuable tool. The ability to

summarize and collect data is useful, to give just one

example. The arrival of AI is, it should be said, also a

reason why the grunt work that junior lawyers used to do is being

replaced. These recent cases are also a reminder that large

AI models, which rely on vast amounts of data, are, in a

way, a sort of “public” field. 

Fintechs are starting to address the problem of how public AI can

be. In

early June, Custodia, a Swiss privacy-first AI startup, said

it had launched Sentinel, a “physical AI thinking appliance

developed and designed from the ground up for executives, family

offices, scientific researchers, and any professional whose

intellectual property is too valuable and too sensitive to trust

to the cloud.”

Perhaps it is not a coincidence that Custodia is Swiss – the land

of bank secrecy (albeit no longer on cross-border matters),

where privacy is still highly prized, as it should

be. 

In its press release about the launch, Custodia said: “Rather

than relying on pre-trained knowledge, Sentinel ingests your

documents, understands their context, and retrieves precisely the

right information to ground every answer it gives. Load thousands

of files – financial records, research papers, legal

documents, corporate history, correspondence – and Sentinel

draws only from that store of knowledge. No hallucination from

unrelated internet data.”

Well, that is the sales pitch, and this news service is looking

into this area about private AI to see just how thick the walls

of privacy really are. What is clear, however, is that if

you are a lawyer or a client, or indeed a professional wealth

manager, investor or professional figure having a confidential

discussion, the AI recording gizmo should be turned off.

Or, at the very least, the use of these devices must be

clearly disclosed ahead of time, giving affected parties a chance

to refuse.