The use of AI in people’s professional and personal lives is becoming second nature. This article highlights some key risks that can arise when AI is employed in the context of seeking and engaging with legal advice. It also suggests how clients may manage these risks.
The comments made in this article are general in nature and application. Your use of AI must always comply with your own organisation’s policies and procedures, as well as the terms and conditions of any relevant contract for the receipt of legal services.
Given how commonplace the use of AI is becoming, an automatic reflex is developing in many people to copy and paste information into AI tools (e.g. to assist in summarising, interpreting and/or to conduct further research on the information).
However, if this is done with material that would ordinarily be covered by legal professional privilege (e.g. legal advice), there is a very real risk that legal professional privilege will be lost over that material.
The problem is that this highest-level of protection – legal professional privilege – only covers confidential communications. Accordingly, if a communication is treated in a way that is inconsistent with maintaining confidentiality, privilege is lost over that communication.
And uploading information to any third party (or online platform) amounts to giving away that confidentiality and privilege.
To ensure the confidentiality of information in the context of AI tools, both operational and contractual terms of service protections must be extremely rigorous.
It is difficult to reconcile that high bar with the act of uploading information onto a public AI platform. Most of these store, collate and replicate uploaded data in some capacity, and may even share data with third parties.
Many AI platforms also utilise servers in foreign jurisdictions, meaning that the treatment of data held on those servers is outside the reach of Australian laws and courts.
On almost a weekly basis, we now receive instructions that show the telltale signs of AI use. They are usually extremely long, follow a similar format, often include “tick box” bullets, and are written in a legalese that is better known to TV script writers than lawyers.
Occasionally, the instructions also inadvertently contain AI chatbot prompts, such as, “would you like help drafting a focused summary of these points?” and, “if you wish, I can draft a short paragraph for inclusion in your briefing to the lawyer.”
This flags a different and serious concern.
Clients in these examples are clearly using AI-generated output to formulate their instructions, but are not consistently or carefully checking them.
Whilst this might get a task off of our client’s desk and onto our desk more quickly, it can result in a chain of easily avoided complexities, and significantly add to the final legal bill.
The problem with unchecked AI use by clients is that it is much more likely to result in instructions and/or draft documents that include:
(a) inaccurate case-specific facts or assumptions;
(b) references to inaccurately stated or non-existent laws; and
(c) incorrect starting premises.
Such unchecked AI use is therefore apt to confuse, create unnecessary complexity and result in additional inputs of lawyer time.
It has always been part of a lawyer’s job to consider whether a client is asking the right question for their needs. However, the factors listed above introduce additional complexity to that task, as a greater amount of work is required to identify, clarify and correct for these factors.
It would be less expensive for clients to spend slightly more time checking AI -generated instructions and documents than to pay a lawyer’s hourly rate to interpret and untangle instructions and documents that suffer from AI-generated deficiencies.
There is also a risk that deficiencies in instructions are not identified until the legal analysis has been completed. In those cases, analysis may need to be re-run (at additional cost) to be fit for purpose.
The risks outlined above can also be exacerbated by the confident tone by which AI chatbots deliver their outputs.
This polish can disguise ambiguities or areas of client uncertainty that may be important for lawyers to interrogate in order to most usefully advise their client.
The usefulness of legal advice often turns on factoring in these areas of grey. A common example of such a grey area is where an approval or authorisation has been received but there are procedural irregularities that may affect its validity.
A similar scenario is where there has been substantial rather than strict compliance with requirements.
Other examples include where there is uncertainty over key facts relevant to legal analysis due to the nature of available records or competing accounts of events.
And whilst AI output often sounds legalistic and authoritative, we see time and again that this is not necessarily reflected in the substance of those outputs.
A common example of this phenomenon is the referencing of non-existent or ‘hallucinated’ case law and legislation by AI chatbots.
There is a real concern that the authoritative tone of AI lulls lay users into a false sense of security and reduces the scrutiny that is applied to AI output.
It goes without saying that the larger the volume of documents a lawyer has to review, the more time-consuming and costly the exercise.
AI-chatbots unfortunately do not have this in ‘mind’ when generating output. They have the capability and propensity to generate large volumes of material very quickly. Far more than a person could realistically generate in a much greater period of time.
Our experience, however, is that this does not necessarily result in more effective documentation. Further, it often results in additional lawyer time being spent verifying the information, correcting the voluminous output and then explaining corrections to the client.
Accordingly, keeping documentation to a ‘human scale’ is an important step to making legal matters manageable and avoiding false economies.
AI is clearly here to stay. It is a critical part of many of our clients’ current and future plans for how they work.
AI should be embraced where it results in genuine efficiencies. We, as a society, are seeing it being used well in many contexts, particularly tasks relating to generating ideas, or organising, summarising and structuring non-sensitive information.
The answer to the potential risks is not to become a technophobe. It is to recognise that some information, for example, privileged legal advice, should never be entered into third-party AI platforms. In addition, to check AI-generated output, especially in legal contexts, and then taken with a grain of salt.
Associate
9200 4900
______________________________________________________________
