In celebration of the upcoming anniversary of this newsletter, we are launching a new series, Jargon Buster. These short, additional newsletters will work through various AI terms, explaining them as clearly as possible (and with a focus on a legal context, where possible). We plan to add a page to the website where all Jargon Buster articles can be searched and explored.

I hope they help anyone puzzled by the new terms appearing every week.

We will start the series with a lighter one, but not one we expect a judge will use anytime soon.

Simply put, a meat proxy is "a human who forwards questions to an AI and forwards the answers back — adding nothing but latency" (see meatproxy.me). I recently encountered meatproxy.me, a humorous exploration of the term with enjoyable artwork, and found myself very fond of the concept. It is something I am encountering more and more in AI use by the general public.

I am confident we have all seen the use it describes. Someone, whether a colleague, a client or a relative, takes a question you asked, inputs it into an AI chatbot, and then repeats the answer back to you, often as if it was their own.

If this is a client relationship, the client is paying for the judgement a lawyer exercises in between those steps, not a regurgitated answer.

This is an AI habit to avoid (and I refer to my published Golden Rules on AI usage). In fact, it may be the AI habit to avoid, going to the very core of bad AI habits.

Along the path of the meat proxy lie all the fears of AI use, expressed by regulators and courts the world over: unverified sources, untested arguments, lazy, unreasoned outputs.

What value has a meat proxy added to the interaction? Have they used any of their own experience and knowledge or understanding in ascertaining the answer? No, despite what I am sure is ample professional experience in their field, the meat proxy has taken no effort to think about the question and simply passed off the answer. Of course, the questioner could have simply asked the chatbot themselves. Presumably, they did not because they valued the input of the meat proxy.

To reference myself again, Golden Rules rule 5 (never rely on an output you have not checked): remains key. Consider the recent case of Cork v Smith [2026] EWHC 1199 (Ch) (National Archives). The AI specifically told the user to verify its answer, and the answer still went to the court unchecked. That is meat proxy behaviour with a published judgment as the receipt.

Consider also Hancox v Sutherland [2026] EAT 139 (GOV.UK), where a litigant in person filed a 300-page ChatGPT skeleton and invited the respondents to find its errors. The EAT's line about "placing the onus to check a document onto an opponent" is the meat proxy in its purest form.

If you are technically minded, classic meat proxy behaviour is giving someone a localhost link. For those that are not as technically minded, localhost is a link meaning "the computer I am using" and therefore sending someone else that link does nothing (a fuller, more eloquent, explanation is here).

There are signs you might be a meat proxy (credit to meatproxy.me):

Whilst the term has a humorous slant, and meatproxy.me even has a feature to refer a meat proxy to their site to educate themselves, the point underneath is serious.

Candidly, I think all frequent users of AI worry that they are becoming a meat proxy. The pull of the shortcut in a busy day is strong. I do not mind admitting that I have found myself reaching for an AI tool when I would perhaps have been better undertaking the task myself.

Since I discovered the term, and reflected on it, I have tried to be more intentional about my AI use, ensuring I am building tools and workflows that support my own skill set, not replace it.

In the legal world, a client is seeking your specific expertise and (given the nature of regulated work) the meat proxy is all the more dangerous and a habit to be avoided at all costs.

A tool should remain simply that: a tool. It is a means to an end of producing your own output. There is value to be gained in workflows that speed up research, organisation and planning, but the output must, in my opinion, remain the human's.

In my view, the best way to avoid being a meat proxy is to question yourself. Can you justify the output you are about to give as your own? You should:

Credit again to meatproxy.me for their excellent site and incredible resources they share freely with the community. The art peppered through this article is theirs, not mine, and I am grateful for the resources.

Artwork: meatproxy.me, used under CC BY 4.0.