Why Local AI Is Becoming Essential for Waterford Solicitors

The confidentiality problem nobody's fully solved yet

Most solicitors have, by now, either tried ChatGPT themselves or watched a colleague use it — for drafting, summarising, or getting a second opinion on a tricky clause. It's useful. It's also a bigger risk than it looks.

The Law Society of Ireland's own guidance on generative AI is unambiguous on this point: free and paid consumer versions of GenAI systems "are not suitable for securely handling personal data or client confidential data." The guidance goes further, warning that sharing privileged communications with a standard GenAI model, without proper safeguards, can amount to intentional disclosure of that material to a third party outside the firm — which risks the loss of legal privilege itself, not just a data protection breach.

That's not a hypothetical compliance footnote. It's the difference between a document staying protected and a document losing its protected status entirely, because it passed through infrastructure the firm doesn't control.

GDPR sits underneath all of this. At a recent Law Society event on AI regulation, the message from panellists was that GDPR "remains fundamental" to how AI is governed in practice, regardless of what new AI-specific rules eventually settle into place — and that firms shouldn't wait for perfect regulatory clarity before acting. The advice was to conduct real, thorough data protection impact assessments now, apply proper vendor due diligence to any AI tool being considered, and treat agentic or autonomous AI features with particular caution, since they can access or process data in ways a solicitor hasn't explicitly authorised.

Picture a fairly ordinary scenario: a conveyancing file with a title document that needs summarising quickly before a call, or a set of discovery documents in a litigation matter that would take a trainee half a day to work through by hand. Dropping those into a consumer AI tool to save the time is an understandable instinct — and it's exactly the situation the Law Society's guidance is warning against. The convenience is real. So is the exposure.

None of this means AI is off the table for a solicitor's practice — quite the opposite, the profession is clearly moving toward using it. It means where the AI runs has become one of the first questions to answer, not an afterthought.

The cost of doing it "properly" keeps climbing — per seat, forever

Assuming a firm takes the Law Society's guidance seriously and looks at AI tools built specifically for legal use, with the security certifications, zero-retention agreements, and data protection terms that consumer tools lack, the cost is real. Legal-specific AI platforms typically run from around €50 to €200 a month per seat for smaller firm tools, and enterprise-grade legal AI platforms with the strongest data protections start at €500 or more per seat, often with annual commitments attached. That's not a one-off cost — it's a recurring line item that multiplies by every solicitor and paralegal who needs access, indefinitely, for as long as the firm uses the tool.

Underlying model pricing tells a more complicated story than "AI is getting more expensive across the board" — budget-tier model costs have actually fallen sharply through 2026, while the most capable "frontier" tiers have largely held their pricing rather than dropping. But that nuance doesn't help a solicitor much, because the tools built for legal use, with the compliance features a firm actually needs, sit in the pricier bracket regardless of what the raw underlying model costs — you're paying for the security wrapper as much as the AI itself.

A local system flips that structure. The cost is the hardware and the setup, once. After that, there's no per-seat licence, no monthly bill that grows every time someone new joins the firm, and no vendor who can change the terms, raise the price, or alter the data policy on you at renewal time. It's a capital cost instead of a permanent operating one.

To put a rough shape on it: a small firm of, say, five fee earners paying even the lower end of enterprise legal-AI pricing would be looking at somewhere in the region of €30,000 a year in subscription fees alone, before any of the underlying model costs the vendor is passing through. A one-time local setup is a fraction of a single year of that, and it doesn't reappear as a bill every January.

Your documents never leave the building — and never need to

This is the part that tends to land hardest once people actually see it: a properly set up local AI system needs no internet connection at all to function. The model runs entirely on local hardware. When a document is uploaded and a question is asked, nothing is transmitted anywhere — not to a cloud server, not to a third-party processor, not across a border. It's genuinely possible to disconnect the machine from the internet entirely and watch it keep working exactly as before.

For a solicitor, that's not a technical curiosity, it's the whole point. It sidesteps the vendor due diligence, the data processing agreements, the "where is this hosted and under what jurisdiction" questions that come with any cloud-based tool, because there's no data transfer happening in the first place. The GDPR and privilege concerns the Law Society has flagged around consumer AI tools simply don't arise in the same way when the processing never leaves the firm's own premises.

It's also the easiest thing to demonstrate convincingly, rather than just claim. In a live demo, it's possible to disconnect the room from the internet entirely — no WiFi, no broadband, nothing — and carry on working with the system exactly as before, because it was never depending on that connection to begin with. Seeing that, rather than being told it, tends to be the moment the confidentiality argument stops being abstract.

Worth being upfront about the other side of this too: a local system isn't a replacement for a solicitor's own judgement, and it shouldn't be treated as authoritative on legal questions any more than a cloud AI tool should. What it's well suited to is the same kind of drafting, summarising, and document-review support firms are already using cloud AI for — just without the data leaving the building to get it.

Why now, specifically

Here's the part that doesn't get talked about enough: the free, downloadable ("open-weight") AI models that make local, offline systems like this possible are currently the subject of a live policy fight in the United States, and the outcome isn't decided yet.

Through the summer of 2026, OpenAI and Anthropic have been pushing Washington toward tighter restrictions on open-weight AI models, framing the issue around national security and concerns about Chinese-developed open models. On the other side, a group that includes Meta, Microsoft, Nvidia, Hugging Face, and Mistral has publicly urged policymakers against what they call "premature restrictions," arguing open models are a genuine security and competitiveness benefit, not just a risk. Notably, the closed-source labs pushing for restriction are also the companies whose subscription-based business models are most directly challenged by free, locally-run alternatives — it's worth taking that into account when weighing the two sides of the argument. As of late July 2026, the White House was reportedly weighing measures including possible bans on specific foreign open-weight models and sanctions, as part of a broader response to AI competition with China.

Nothing about locally-run AI disappears if new restrictions land — models already downloaded keep working regardless of future policy. But the range of options available, and how easily new and better open models can be adopted down the line, is genuinely being contested right now by companies with a direct commercial interest in that not happening. Getting set up on a local system while the current generation of strong open-weight models remains freely available is a reasonable, and reasonably urgent, thing to act on rather than leave for next year.

See it for yourself — at your own office, with your own documents

The easiest way to know whether this is right for your firm isn't to take my word for it, it's to watch it work in your own building, on your own machines, with your own eyes on the screen.

I'm based in Portlaw, and I'm offering a free, no-obligation demo day at your office: I bring the equipment, set up a fully offline local network, and let your team put it through its paces on some sample documents. No internet connection in the room, no data going anywhere, no cost, and no commitment — just a straight answer to whether this is worth pursuing further.

If you'd like to see it for yourself, get in touch and we'll find a day that suits.