Guest post by John Hayes, Managing Partner at London employment and regulation firm Constantine Law
Hayes: Are internal summaries of AI-generated legal advice privileged?
Much has been written about the risks of lawyers misusing AI for legal research and drafting. Recent cases involving fabricated authorities and inaccurate legal analysis have rightly garnered attention.
But in my view, the bigger, more pressing challenges lie elsewhere. It is the everyday use of AI by the clients themselves.
Every day, clients upload legal advice, correspondence, and draft documents to AI platforms like ChatGPT and Claude.
They do so for understandable reasons, such as to criticize legal advice, test its accuracy, summarize it for colleagues, or make it more accessible to decision makers within an organization. But doing so can put them at significant risk to themselves.
Consider our recent example. One client responded to a draft of a letter one of the partners wrote with the following comment: “Congratulations, you got 19 out of 20.”
When I asked what that meant, my client explained that he had uploaded a letter to an AI platform and it received a 19 out of 20 score for accuracy.
An interesting discussion ensued. Our partners were ultimately able to prove that the AI model failed to recognize the nuances of the key facts raised in the letter. In other words, the missing points were not the lawyer’s mistake, but the AI system’s own mistake.
On one level, this is a funny story. In another aspect, it marks a significant change. For the first time, lawyers can expect their work to be regularly reviewed, scored, and challenged by clients equipped with increasingly sophisticated AI tools.
The democratization of legal knowledge has already begun. AI has given non-lawyers unprecedented access to legal analysis and reasoning, enabling many to act as relatively effective first instance advisors.
The impact is already visible. As an employment law practice, we now routinely receive 15-20 page draft complaints and grievances prepared with the assistance of AI. A few years ago, equivalent documents were often four to five pages long.
Ironically, AI is not currently reducing the amount of legal responses. It is increasing.
However, the expansion of legal red tape is not the main concern.
A more serious issue is the impact of clients’ use of AI on confidentiality and legal privilege.
Clients are increasingly uploading legal advice to AI systems for the purpose of generating internal summaries.
In one recent case, a client uploaded a detailed letter from Constantine Law to Claude and asked the system to create a short summary that could be forwarded to his line manager. The summary was then circulated internally.
This seemingly innocuous behavior raises two important concerns.
First, sensitive and legally privileged communications may be disclosed to third-party AI platforms. Legal consequences will vary depending on the situation, the platform used, its terms of service, and applicable law.
However, clients should not assume that uploading privileged content to AI systems is without risk. Any potential waiver of confidentiality or privilege must be taken seriously.
Second, once an AI-generated summary is created and distributed, the question arises as to whether the summary itself attracts legal privilege. A summary is no longer an attorney’s correspondence. It is a new document generated by the interaction between the user and the AI system. Whether a document is privileged or not may not be straightforward.
In highly sensitive and contentious disputes, which are the case in many employment cases, these distinctions are very important. If privilege is lost or successfully challenged, communications that the client assumed were protected may be disclosed in litigation.
Internal interactions between managers discussing summaries of AI-generated legal advice can be fertile ground for disclosure requests.
The danger is that clients may unwittingly undermine one of the most valuable protections the legal system provides.
There are no easy solutions. The reality is that AI is now embedded in the workplace. Clients continue to use these tools to analyze, challenge, and summarize legal advice. That genie won’t go back in the bottle.
Privilege law has traditionally been narrowly interpreted and is easily undermined by the careless dissemination of legal advice. Whether this law evolves to accommodate the realities of AI-assisted communications remains an open question.
Courts and legislators may ultimately develop new principles that reflect modern business practices.
But so far, the law has not fully caught up with the practical realities of AI.
Clients should therefore understand that the unrestricted use of AI tools in connection with legal advice carries real risks. In some circumstances, confidentiality may be jeopardized, privilege claims weakened, and one of the main benefits of seeking legal advice in the first place.
There is still much to be discussed and resolved. But from my perspective, the most significant AI challenge facing the legal profession is not the use of AI by lawyers. That is the use of AI by clients.
