Crossing the rubicon
AI has reached a ‘Rubicon moment’ for the legal profession. The question is no longer whether to cross it – but who goes first, and whether those who hesitate will find themselves stranded on the wrong bank? Sorcha Corcoran reports from the AI in Legal Practice Summit.
The main thread running through the recent AI in Legal Practice Summit in Blackhall Place was the sheer pace at which artificial intelligence (AI) is advancing, making it imperative for everyone in the legal profession to get to grips with it – and to figure out what it means for them and their law firm.
On the same day, 26 March, members of the European Parliament agreed to postpone obligations for certain high-risk AI systems under the EU AI Act as part of the ongoing debate of the Digital Omnibus Package – a legislative initiative published in November aimed at simplifying digital laws.
The European Parliament’s position is that the compliance deadline for standalone high-risk AI systems (Annex III) should be pushed out from 2 August 2026 to 2 December 2027, while the one for high-risk AI embedded in products (Annex I) should kick in on 2 August 2028, instead of 2 August 2027.
Delays negotiation
Having confirmed the text of the Digital Omnibus Package, the European Parliament will now have to negotiate the proposed delays with the European Council and the European Commission.
These are the latest deadlines in the phased implementation of the EU AI Act – officially in place since August 2024 – to be either missed or possibly moved as the legislator continues to try to play catch-up with the technology.
“The proposal for the EU AI Act was first published by the European Commission in 2021, and the AI that initial piece of legislation was dealing with was completely different to generative AI [genAI]. Other than early research versions, there was nothing in the public eye like the large language models and transformer architectures that we have now,” said Dr Barry Scannell (partner at William Fry), who chaired the AI in Legal Practice Summit.
At present, the European Parliament is trying to expand the list of prohibited practices under the act, which became legally binding in February 2025, to include nudification-type apps. This follows the Grok scandal, where the X chatbot was generating non-consensual, sexualised deepfake images.
“We’re talking about a legal framework where you’re trying to make laws for things that were literally science fiction ten years ago,” Scannell observed.
Dramatic impact
In his view, AI is going to have a dramatic impact on the legal profession here, and soon.
“In some jurisdictions, there are already significant impacts happening in areas such as graduate hiring, in the number of people being kept on after training, and so on,” he said.
“And AI really is changing the structure of law firms and how we do work in Ireland – even fees are being affected. This is not something that might happen in the future; it’s starting to happen now.”
During the panel discussion, an anecdote shared by Kyle Gribben (head of Matheson’s Digital Services Group) about his experience of a recent roundtable breakfast meeting in London drove this point home.
A person at the meeting questioned why he should have to pay a law firm to do certain parts of his work when he had a choice of four different AI platforms at his disposal, in-house, that could do it for him.
He stated that he wanted to give every law firm on his panel 12 months to explain how they were going to cut his fees by 50% using genAI.
Billable-hour threat
“The billable-hour model has been written off on several occasions over the past 20 years. I don’t think it’s going to go away lightly or easily. But this is the one time that it’s probably really, actually under threat,” said Gribben.
“That person at the breakfast wasn’t the first I’ve heard speaking like that about it. But they’re not taking into account the value they’re getting when a lawyer is using AI – that it costs money to bring in these platforms and train people up on them.
"There’s also the value in ensuring equality of arms by using AI when the other side is represented by someone who is able to enhance their practice with these technologies.
“So, it’s not just a simple thing of less time equals less value for what the client gets. I think what we’re probably going to see is some firms taking the first step away from the billable hour. There has to be someone that crosses the Rubicon, and when they do, they’ll be watched very intently by many people all across the world.”
Human in the loop
Regardless of what happens with how work is priced, Gribben maintains that AI doesn’t change what clients are actually buying from lawyers.
With this in mind, he advised practitioners to see it simply as a new toolkit to practise with, and to see themselves as professional users of genAI tools.
“The key to all of this is to understand that, no matter what you do with AI, no matter what sort of efficiency you get from using it, you must make provision for a human in the loop.
"So, when pricing for a client, you might have an AI tool that can review a piece of work in 20 minutes rather than four hours, but you still have to build in time for you to review the output afterwards and be that professional user,” he said.
In his keynote address, technology futurist and author David Kerrigan cited a LexisNexis survey published in September that found legal leaders to be still struggling to balance speed with accuracy, increase billables without extra workload, and deliver greater value without undermining legal expertise.
It showed that 61% of UK lawyers were using genAI in their work – up from 46% eight months previously.
AI tightrope
Kerrigan said: “Most people I come across regard AI as a tightrope. It’s not an easy path to navigate. When properly applied, it can deliver real business benefits today.
"But there are concerns about what AI is going to do to the legal profession: will it dumb things down? Will it make mistakes? Will it expose you as a practitioner? These are genuine and well-founded concerns.”
In the past six to 12 months, he has noticed that the tone of the debate around AI has become a lot more strident: “We’ve passed the phase of having polite discussions about whether or not to use it. Bosses have been warning employees to use AI or face the sack, and that not being on board with it will affect their chances of promotion,” he said.
“I always point out that the AI we now know – genAI – is still very young and nowhere near the potential that it will have in the future. It will be very different as we go forward. So an important decision point for legal practitioners is, when is it good enough to be useful?”
Just another technology?
When it comes to adoption, one of the common mistakes Kerrigan has observed all over the world, in all types of industries, is to think of AI as just another technology.
“It’s incredible the number of companies that think it’s the equivalent of a new version of Microsoft PowerPoint – that they can just deploy it and it will work. It’s not like that,” he said.
“It’s also not one thing. AI means different things to different people. How hard it is to define is one of the reasons why legislating around AI is so difficult. It’s not a simple, straightforward concept and absolutely not something that you can hand over to your IT department to fix something.
“The reality about AI is that it’s a mindset problem – it’s all about how you approach your work to leverage the tools. So implementing it starts with the business process you’re trying to improve, not with the AI itself. You have to arrange how you think differently to make the most of what AI is capable of.”
LESSONS FROM A COURTROOM
At the summit, Ann Henry (partner at Bird & Bird) highlighted how a recent American ruling on AI-generated documents has significant implications for Irish practitioners. It’s a stark warning for clients who treat chatbots as lawyers.
“A US federal court has issued what appears to be a first-of-its-kind ruling on the status of AI-generated documents in criminal proceedings, with consequences that extend well beyond American borders,” she said.
“The judgment of Mr Justice Rakoff on 17 February 2026 in United States v Heppner raises urgent questions about attorney/client privilege, confidentiality, and the governance of AI tools in legal practice.”
The defendant, Bradley Heppner, was indicted on fraud-related charges, and the FBI executed a search warrant at his home, seizing numerous documents and electronic devices.
Knowing some months previously that he was being investigated, Heppner used Anthropic’s AI platform Claude to prepare reports outlining his own defence strategy – without any direction from a lawyer.
Following his arrest, about 31 AI-generated documents were recovered from seized devices. The government sought a ruling that these documents did not attract attorney/client privilege.
Judge Rakoff agreed on three grounds, Ann Henry explained:
“First, Claude is not a lawyer, and there was, therefore, no communication between client and counsel.
“Second, confidentiality – the bedrock of privilege – had been destroyed by Anthropic’s own privacy policy, which permitted the collection of users’ inputs [prompts] and their outputs, that it then used to train the AI model. As such, it reserves the right to make disclosure to third parties, including governmental regulatory authorities, in certain circumstances.
“Third, the fact that Heppner later shared the documents with his lawyer did not retroactively clothe them with privilege.”
Henry said that it was more likely than not that an Irish court would reach the same conclusion. This is because the conditions for privilege under Irish law touch upon those same issues considered by the US court, namely:
1) Is there a communication between a client and lawyer?
2) Was it made in confidence?
3) Was it made in the course of a professional legal relationship? and
4) Was the communication for the purpose of the giving or receiving of legal advice?
“The Heppner case is a lesson on the importance of using AI tools with enterprise-grade terms and conditions that ringfence confidentiality for users,” Henry advised.
Governance concerns do not end with privilege, she added, highlighting a separate case, Nippon Life v OpenAI, Northern District Illinois, which was filed on 4 March 2026.
Nippon Life has issued proceedings against OpenAI, alleging tortious interference, the unlicensed practice of law, and abuse of process arising from ChatGPT allegedly inducing a user to reopen a full and final settlement that had been reached with Nippon Life.
“That case illustrates a broader risk – that of clients increasingly treating AI tools as a means of obtaining legal advice with potentially detrimental consequences. This issue creates challenges for clients, lawyers and, importantly, society,” said Henry.
Sorcha Corcoran is a freelance journalist and a regular contributor to the Gazette.