The Business of Law · Cyprus

Artificial intelligence is compressing the cost of producing legal work. The more useful question for a Cyprus firm is not how to defend the work, but how to defend the reason the work exists.

By Philippos Aristotelous · 16 min read · Last reviewed 7 September 2026

In March, an advocate in Limassol called Elena Georgiou put a bulk carrier on the Cyprus register. Her client, a shipowner in Piraeus, had bought a fourteen-year-old vessel and wanted it under the Cyprus flag before the end of the quarter. Elena has done this many times. The difference this year was that the first drafts of the documents took her about forty minutes rather than most of a day, because she now works with an AI assistant and has stopped pretending otherwise.

She told me she was not especially worried about that. What she was worried about was the sentence underneath it. Her fee for that registration has been roughly the same for six years, and if the drafting keeps getting quicker, it will not stay there. But the fee is not really the point either. The point is that the whole engagement existed because a Greek shipowner decided that a Cyprus-flagged vessel was worth having. Nothing in her software made that decision, and nothing in her software can.

That is the argument I want to make here, and I want to make it carefully, because it is easily mistaken for a comforting one. It is not.

ONEWhat AI is actually making cheap

Start with what the evidence supports, because a great deal of what is written about AI and law is either sales material or anxiety.

The most useful studies are the randomised ones. An earlier trial at Minnesota gave sixty law students access to GPT-4 across a set of legal tasks. Speed improved by something in the order of eleven to thirty-two per cent depending on the task; quality improved only slightly and inconsistently.¹ That was the honest state of things when GPT-4 was the best available tool, and it was a good deal less dramatic than the headlines of the period.

The follow-up study, first reported in 2025, is more striking. A team led by Daniel Schwarcz ran a randomised controlled trial at Minnesota and Michigan, enrolling 153 law students, of whom 125 completed all six tasks, and comparing no AI against a reasoning model and against a retrieval-based legal research tool. This time both quality and speed moved. Productivity rose significantly on five of the six tasks — by roughly fifty to a hundred and thirty per cent, depending on the tool — and quality improved too, though far more modestly: by between about a quarter of a point and a point and a quarter on a seven-point scale.²

Two details in that study matter more than the headline. The first is that the transactional drafting exercise was among the tasks on which neither tool produced a measurable quality gain — quality improved significantly on four of the six. The second is that the tools still produced fabricated material, and the reasoning model produced more of it than the retrieval-based one — eleven instances against three, on counts the authors present as descriptive rather than statistically significant.

Alongside that sits the reliability question. The Stanford RegLab team tested the leading commercial legal research tools and found that they produced hallucinated output in something between seventeen and thirty-four per cent of queries: seventeen per cent for Lexis+ AI and for Ask Practical Law AI, thirty-four per cent for Westlaw’s AI-Assisted Research.³ Better than a general chatbot, and nowhere near good enough to be left unsupervised. The consequence is now visible on the court record rather than merely arguable: a database maintained by Damien Charlotin had logged 1,641 decisions worldwide dealing with AI-fabricated citations by 24 June 2026, and an analysis of an earlier snapshot found that roughly thirty per cent of them attracted a sanction.⁴

Then there is adoption, which is a different thing again and is usually reported as if it were the same. Thomson Reuters surveys professionals each year and reports that they expect to save around five hours a week. That is a forecast made by respondents about themselves, not a measurement of anything. The same survey found that only about a quarter of organisations had a visible AI strategy.⁵ I read that as the more informative number.

None of this tells us that lawyers are being replaced. It tells us something more specific and, commercially, more awkward. The scarcity is moving. For most of the profession’s history, the ability to produce a competent legal document quickly was itself scarce, and firms were paid partly for that scarcity. That particular scarcity is thinning.

TWOThe question beneath the AI question

Which brings me to the question I think is being skipped.

Almost every conversation I have with managing partners about AI is a conversation about the work: which tasks it can do, what we should let juniors use, how to stop it inventing authorities, whether we should still bill by the hour. All reasonable questions. But they all sit at one layer, and there is another layer underneath.

Why does an international client need Cyprus counsel at all?

Because an asset is here. Because a company was incorporated here. Because a vessel flies this flag. Because Cyprus law governs the agreement. Because a regulator here requires something. Because a transaction touches this jurisdiction. Because somebody must take legal responsibility for an act, in a place, under a professional obligation that can be enforced.

Every one of those reasons is anchored in something physical, institutional or legal. A model can help a lawyer discharge the resulting work far more quickly than before. It cannot manufacture the anchor.

THREEOur advantage was never the drafting

Cyprus lawyers have tended to describe their advantage in terms of craft: we know the law, we draft well, we are responsive, we understand international structures. All true, and all increasingly difficult to charge a premium for, because a competent lawyer anywhere with good tools is closing the gap on the production side.

The advantage was always somewhere else. It was that a great deal of economic activity chose to sit here, and that Cyprus law and Cyprus institutions were attached to it. An EU member state with a common-law inheritance and English-language practice. An extensive treaty network. A corporate registry that international groups understand. A funds and financial services regime under an EU regulator. A shipping cluster with genuine operational depth in Limassol, not just a letterbox. Courts, arbitration, and a professional ecosystem of lawyers, accountants, administrators and bankers who have worked together for decades.

Those things generated the work. The firms captured part of it. That is not a criticism — it is how every professional-services centre in the world operates. It is only dangerous if we forget it, and mistake our share of the demand for the source of it.

FOURThe gate and the traffic

Here is where the comfortable answer usually arrives, and I want to deal with it properly, because it is the answer I hear most often and it is half right.

The Advocates Law, Cap. 2, defines what it means to practise as an advocate, and section 11 provides that no person shall practise as an advocate unless enrolled, holding a current annual licence and up to date with the Advocates Pension Fund.⁶ The definition in section 2 is broader than many people outside the profession realise. It covers, among other things, appearing before a court and preparing pleadings, and it also covers, in the words of the statute, the drawing, reviewing or amending of any memorandum or articles of association of a company of any form, and of any document pertaining to the incorporation, registration, organisation, reorganisation or dissolution of any legal entity. It covers registering ships and drawing all documents referring to the creation, transfer, alteration or abolition of rights on a ship. And it covers giving opinions on legal matters submitted to the advocate.⁷

The statute goes further on company constitutions specifically. Section 11(5) makes the drafting of a memorandum or articles the exclusive work of a practising advocate, who signs and seals them, and provides that a memorandum or articles drafted in breach of that requirement is void and of no legal effect. There are criminal penalties attached.⁸

Statutory audit sits behind a comparable gate: it may only be carried out by auditors and audit firms approved and registered under the Auditors Law of 2017, supervised by the Cyprus Public Audit Oversight Board.⁹

The professional overlay is moving too, though more slowly than the technology. The Council of Bars and Law Societies of Europe published a guide on the use of generative AI by lawyers in October 2025, and the Cyprus Bar Association has circulated a Greek-language synopsis of it to the profession.¹⁰ It is guidance rather than a binding Cyprus rule, and it changes nothing about the reserved activities. But it is the point at which the duties of competence, confidentiality and supervision begin to be written down for advocates who use these tools, and it is worth reading before a firm discovers the question in a complaint.

So it is quite true that a language model cannot incorporate a Cyprus company, register a vessel, or sign an audit opinion. The gates are real, they are statutory, and they are not going to be dissolved by better software.

And that is the whole difficulty. Those provisions were drawn in 1981, 1984 and 2007. They were written to regulate who may act, not to guarantee that anyone would want the act performed. They protect a step in a process. They do not protect the existence of the process.

Consider how that erodes without a single amendment to the Advocates Law. A group decides that its holding company will sit in another member state, so the incorporation never happens here. A shipowner registers elsewhere, so the ship registration document is never drawn. A client automates everything around the reserved step, compresses the professional input to the narrowest act the statute actually requires, and pays for twenty minutes instead of six hours. Or the reserved work is performed slowly and expensively enough that the client’s next structure is designed to need less of it.

In each case the monopoly is untouched and the revenue is gone. This is the part I would ask managing partners to sit with. Reserved activity is a floor, not a moat. The moat, if a Cyprus firm has one, is the jurisdiction itself — and moats are maintained, not inherited.

FIVECyprus is already being tested — in both directions

It would be easy to write the next section as a warning. The honest position is more mixed than that, and the mixture is the interesting part.

From 1 January 2026 the corporate income tax rate rose from 12.5 per cent to 15 per cent.¹¹ That is a real change, and it has been widely read as the end of a long-standing advantage. I think that reading is too quick. Cyprus had already legislated the EU global minimum tax rules in December 2024, which impose a fifteen per cent effective floor on groups with consolidated revenue above €750 million.¹² For those groups, fifteen per cent was already the practical position; the headline move mostly aligns the domestic rate with it. The genuine increase falls on the smaller and mid-sized international groups that sit below that threshold — which happens to describe a large share of the client base of Cyprus law firms. What is less often said is that the same December 2025 package moved several things in the opposite direction. The deemed dividend distribution regime was abolished for profits arising from 2026, the special defence contribution on dividends paid to domiciled individuals fell from seventeen per cent to five, and the Stamp Duty Law was repealed outright.¹³ Those measures take real friction out of exactly the structures that carry the rate increase. The net position for that client base is genuinely mixed, and a firm that describes it to clients as simply worse is getting it wrong in a way the client can check.

Meanwhile the shipping sector went the other way, and it did so on foundations the tax change did not touch, since tonnage tax operates outside the corporate income tax and the European Commission has approved the Cyprus regime to the end of 2029.¹⁴ At the ninth Capital Link Cyprus Shipping Forum in Limassol on 17 February 2026, the Shipping Deputy Minister, Marina Hadjimanolis, said the registry had recorded a twenty-three per cent increase in gross tonnage over the preceding two and a half years, its highest level in twenty-five years.¹⁵ By August 2026 government figures placed Cyprus third in Europe and eleventh in the world by registered fleet.¹⁶ Ship management is now the larger economic story: Central Bank of Cyprus survey data put sector revenue at €978 million in the first half of 2025, equal to 5.5 per cent of GDP,¹⁷ and at about €1 billion in the second, equal to 5.4 per cent, taking the year to roughly €1.98 billion.¹⁸

I should correct something here that I have said myself in shorter form. The right figure for ship management is around 5.4 to 5.5 per cent of GDP, not more than 5.5 per cent. The number is impressive enough without rounding it upwards.

Company formation also moved in the encouraging direction. Registrar of Companies figures show 18,858 new company registrations in 2025 against 14,444 in 2024, an increase of just over thirty per cent, with net registrations after strike-offs and dissolutions rising from 3,738 to 5,743.¹⁹ The comparison has to be drawn on a single basis to mean anything: the Registrar also publishes a wider series counting all new business entities, which takes in partnerships, business names and overseas companies as well, and that broader figure was 14,908 for 2024. It is the narrower company count that the net numbers are built from.

Against that, the compliance environment has become materially heavier. Successive EU sanctions packages have removed a body of business that Cyprus firms once serviced, and the reach is still widening: the twentieth package, adopted on 23 April 2026, extended the Article 5n prohibition — which already caught legal advisory, accounting, tax and business consultancy services supplied to entities established in Russia — to managed security services as well.²⁰ Cyprus criminalised the violation of EU restrictive measures through Law 149(I) of 2025 and, by Law 150(I) of 2025, created a National Sanctions Implementation Unit inside the Ministry of Finance with powers to license, investigate and fine.²¹ Screening, verification and refusal now consume time that used to be billable advisory work, and getting it wrong has become a criminal exposure rather than a supervisory one. That is not a small adjustment to a professional-services economy.

Read together, these are not the numbers of a jurisdiction in decline. The economy grew 3.8 per cent in real terms in 2025 and is forecast to moderate to somewhere between 2.3 and 2.6 per cent in 2026 — a slowdown, not a reversal.²² They are the numbers of a jurisdiction whose demand is being re-sorted: strengthening where there is real operational substance, weakening where the proposition was mainly a tax rate and a company secretary.

SIXWhat Delaware has understood for a century

This is not a Cyprus phenomenon, and it is worth one comparison to show that.

Delaware is the clearest case in the world of a jurisdiction that is itself the product. Corporate franchise taxes together with the associated entity fees generate somewhere between a quarter and thirty per cent of the state’s general fund revenue; the franchise tax on its own raised about $1.3 billion in the 2025 fiscal year.²³ An enormous corporate bar exists there for one reason: companies choose to be Delaware companies.

When that choice began to look less automatic — a run of high-profile reincorporations elsewhere, and a good deal of noise about it — Delaware did not respond by tightening the rules on who may practise law. It amended its corporation statute in March 2025, through Senate Bill 21, to restore the predictability its customers said they had lost.²⁴

What happened afterwards is the more instructive half of the story. The amendment did not stop the outflow. Boards put twenty-six reincorporation proposals to public-company shareholders during 2025, and something in the order of forty-nine such proposals were filed between January 2024 and March 2026, with Nevada and Texas taking most of the traffic.²⁵ Delaware’s revenue has held up so far. But a jurisdiction that legislated specifically to keep its customers found that it then had to keep competing for them, which is rather the point.

That is the instinct worth borrowing. When a jurisdiction competes for demand, the profession that depends on it competes too, whether or not it participates in the conversation.

SEVENTwo strategies, held at the same time

What follows is my own view rather than anything the evidence compels, and I would put it as two obligations that have to be carried together.

The first is jurisdictional, and most firms behave as though it were somebody else’s job. It is not. Cyprus law firms have more influence over the quality of this jurisdiction than almost any other private constituency: they see where structures fail, which procedures take too long, which registries frustrate clients, where the reputational damage is being done, and which reforms would actually matter. Most of that intelligence never leaves the building. Firms that are serious about the next decade should be putting it in front of government, the Bar, and the sector bodies, and should be as interested in whether Cyprus wins the next mandate as in whether they do.

The second is internal, and it is not an AI checklist. It is four questions asked honestly. Where does our demand actually originate — which structural feature of this jurisdiction produces it, and is that feature strengthening or weakening? Which of our work is becoming commoditised, and are we still pricing it as though it were scarce? Where does professional accountability genuinely carry value, as against merely being required? And where can AI take real cost and delay out of what we do, so that the client’s reason for being here is reinforced rather than tested?

I have written elsewhere about the pricing consequences of this, about what the tools can practically do inside a Cyprus firm, and about the implementation gap in the national AI strategy. This piece sits underneath all three. None of that work matters much if the demand is drifting somewhere else while we optimise how we serve it.

IN CLOSINGThe registration, and the reason for it

Elena will register more vessels this year than last, and she will draft the documents faster than she ever has. Both of those things are true at once, and only one of them is within her control.

The profession has spent a great deal of the past two years asking whether AI can perform today’s legal work. It is the wrong end of the question. The work exists because something real is here — a ship, a company, an asset, an obligation, a regulator, a signature that somebody must be answerable for. Defend that, and the rest is a question of how well you use the tools.