THE AUTHOR:
Nikolaos Kolotouros, Global Partnerships and Legal Publications Intern at Jus Mundi
On 15 October 2026, Jus Mundi and the International Chamber of Commerce (“ICC”) hosted the panel “AI Built on Standards” at Jus Mundi’s offices in Paris, bringing together representatives from the legal, technology, legal information and arbitration sectors to examine how organisations can create sustainable and responsible value with Artificial Intelligence (“AI”).
The panel was moderated by Arvydas Ziobakas, AI & Data Strategy Global Lead at the ICC. The panellists included Sébastien Bardou, Managing Director for Northern and Southern Europe and Western and Central Africa at LexisNexis; Juan Carretero Sánchez, Head of Legal at Microsoft Portugal; Ayushman Dash, Head of AI at Jus Mundi; Claire Dilé, Public Policy Specialist at Anthropic; Muriel Dossi Gbegnito, Senior Innovation Advisor at Linklaters Paris; and Giovanni Ricciolio, Information Systems Director at ICC Arbitration.
The discussion focused on the practical challenges of AI adoption in the legal sector, moving from questions of value and productivity to organisational transformation, quality assurance, trust, regulation and AI fluency.
From Widespread Adoption to Measurable Value
The panel opened by considering whether AI adoption in the legal industry is already generating measurable value or remains primarily a matter of enthusiasm and experimentation.
Figures cited during the discussion suggested that AI adoption is now widespread across the legal sector, with approximately 95% of legal organisations reporting that they use generative AI. However, the number of organisations able to demonstrate a tangible impact on the quality of legal work remains significantly lower.
According to Sébastien Bardou, the first wave of legal AI adoption focused primarily on research. More than 80% of early usage of LexisNexis’s AI tools related to what the company describes as the “ask” use case, namely using AI as a more sophisticated and contextualised research engine.
The focus is now shifting towards agentic AI, drafting, workflow automation and the use of an organisation’s own internal data. This represents a significant change from simply asking an AI system to retrieve information. Legal professionals are increasingly looking to build repeatable workflows and specialised skills that can process both authoritative legal content and internal organisational knowledge.
The panellists nevertheless stressed that productivity alone should not be treated as the primary measure of value. Juan Carretero Sánchez illustrated this point with a practical example: a task that previously required one hour can now be completed in just thirty minutes with AI assistance. The real question, he noted, is what happens with the thirty minutes saved.
If that time is simply redirected towards more low-value administrative work, the organisation may not have created meaningful value. By contrast, if it is used to improve work-life balance, engage more proactively with internal clients or focus on more strategic legal issues, the benefit becomes much more significant.
This human-centred approach also challenges the assumption that AI will replace legal professionals. The discussion repeatedly returned to the distinction between replacing tasks and replacing jobs. AI may automate or enhance specific tasks, but the panellists observed that there is currently no evidence of large-scale job replacement across the legal sectors of the United States and Europe.
Better Outcomes for Clients, Not Simply Lower Fees
The conversation then turned to the impact of AI on the end users of legal services.
For in-house legal teams, AI can create additional capacity to engage proactively with business units. Rather than waiting for a legal issue to arise, legal teams can participate earlier in strategic discussions, identify risks and help shape business decisions.
Carretero Sánchez explained that AI has enabled in-house lawyers to respond more quickly while also handling a larger number of questions from business teams. In this sense, AI is not necessarily reducing the demand for legal advice. Instead, better-informed business teams may generate more sophisticated questions, increasing the need for high-quality legal input.
The same dynamic may affect the relationship between law firms and their clients. Clients are not necessarily expecting legal fees to fall simply because AI is being used. Instead, they may expect greater depth, speed, and quality.
Sébastien Bardou gave the example of due diligence in an M&A transaction. Traditionally, a law firm might focus on a limited number of key contracts and review a sample of smaller agreements. AI can analyse a much larger body of contracts, identify inconsistencies, and uncover risks that might otherwise remain hidden in less prominent documents.
The value of AI, therefore, may lie less in reducing the price of legal services than in enabling lawyers to provide a more complete service. This may also accelerate the shift away from purely hourly billing towards models that reflect the advice, outcome and value delivered to the client.
The panellists also noted that transparency about the use of AI is becoming increasingly important. Law firms and legal departments are not only using AI internally, but are also communicating more openly with clients about the tools and processes involved. Responsible use of AI can therefore become a competitive advantage rather than a source of concern.
AI Adoption Is an Organisational Challenge
One of the clearest points of consensus was that successful AI adoption depends less on selecting the latest technology than on preparing the organisation to use it effectively.
Juan Carretero Sánchez identified three central pillars: AI fluency and skills; process redesign before product selection; and knowledge management. Organisations may acquire powerful AI tools, but their value will remain limited if employees do not understand how those tools work, which processes they should support and what limitations they have.
The panellists warned against purchasing AI licences without first identifying the specific workflows that need to be improved. A tool may be technically impressive but unsuitable for a particular process, especially when the task is low-risk or the organisation lacks the internal data needed to support it.
Knowledge management presents a further challenge. Legal organisations often contain significant expertise, but that knowledge may remain fragmented across individuals, teams, documents, email accounts and disconnected systems. AI can help connect people, processes and institutional knowledge, but only if the underlying information is properly organised and maintained.
The discussion also highlighted the importance of clear ownership. If an AI-supported process has no owner, it is unlikely to be properly maintained over time. Governance, data policies, leadership and accountability are therefore essential components of adoption.
Muriel Dossi Gbegnito emphasised the role of experienced lawyers in leading organisational change. Senior practitioners can help establish confidence in new tools and encourage younger lawyers to adopt them responsibly. AI adoption is consequently not only a technology project, but also a leadership and transformation programme.
Trust Is Built Through Technology, Data and Critical Thinking
The panel identified three complementary foundations of trust in legal AI.
The first is the technology itself. Models are becoming more capable, while software layers can now manage permissions, context, memory and access controls. These developments can reduce errors and make AI systems more suitable for professional environments.
The second foundation is the data on which the system is grounded. The panellists distinguished between open web content, which may be incomplete, inconsistent or unreliable for legal research; internal organisational data, which can be highly valuable but requires continuous maintenance; and proprietary legal content, including authoritative and jurisdiction-specific materials produced by legal publishers and institutions.
The third foundation is the critical thinking of the end user. Legal professionals are well placed to use AI because their work requires scrutiny, analysis, and the ability to identify exceptions. However, the panellists cautioned that users could become complacent when interacting with AI, even when they claim not to trust it.
Trust was therefore described as an output rather than an input. It is not created simply by declaring that a system is reliable. It develops through consistent delivery, transparent processes, and the user’s ability to challenge and verify the result.
Quality Assurance and Transparency
The discussion also explored what makes legal AI different from AI used in less sensitive areas.
In legal work, even a small error can have serious consequences. A mistaken case citation or an inaccurate statement of law is not comparable to a minor error in a less consequential context. The tolerance for error is therefore extremely low.
Ayushman Dash explained that quality assurance in legal AI can be understood through two layers. The first concerns coverage and reliability: the system must retrieve the relevant materials, provide accurate citations, comply with legal conventions, and produce answers that are relevant, faithful and sufficiently comprehensive. The second concerns the additional insight that AI can provide beyond a conventional manual search. This “delta” or “wow effect” is more difficult to measure because it requires an assessment of whether the system has identified something that a lawyer might otherwise have missed.
The panellists encouraged organisations adopting legal AI to ask providers detailed questions about their quality assurance processes, including which datasets were used for testing, how large and representative the test cases were, which metrics were applied, how citations and sources were evaluated, and how users can test the system against their own workflows.
Transparency should not stop at the pre-sale stage. Users should also be able to understand how an AI agent reached its result, which databases it consulted and which steps it took along the way.
The discussion referred to Retrieval-Augmented Generation (“RAG”) and agentic systems as examples of how legal AI providers are addressing these challenges. Rather than training a model once on a static legal corpus, systems can retrieve relevant content from an evolving body of legal materials. Agentic systems can then coordinate specialised searches across different types of content, while displaying the steps taken and the sources consulted.
The Future Legal Operating Model
As AI becomes more capable, the legal profession will need to reconsider how expertise is developed.
One question raised during the discussion was how junior lawyers will become senior experts if AI performs many of the tasks traditionally used to develop legal judgment. Future lawyers may need to learn not only how to produce an answer, but also how to determine whether an AI-generated answer is valid in the relevant legal and business context.
This makes transparency and customisation particularly important. Legal professionals must remain able to understand the reasoning behind an output, challenge it and adapt the system to the requirements of a particular matter.
The future operating model is therefore unlikely to consist simply of a small number of people supervising autonomous agents. It will require professionals with sufficient substantive expertise to evaluate AI outputs, identify the significance of the remaining margin of error and retain responsibility for the final decision.
Regulation, Sovereignty and Global Standards
The final part of the discussion addressed regulation, including the EU AI Act 2024.
The panellists broadly welcomed the EU framework, describing it as flexible and capable of adapting to technological developments. At the same time, they noted that agentic AI is likely to fall within the high-risk category in certain legal contexts, while research-based use of AI would not necessarily be high-risk in itself.
The main concern identified was the documentation and compliance burden that high-risk classification could create, particularly for courts and judicial systems that may already lack resources.
The panel also stressed that the EU AI Act is regional, while AI development and deployment are global. Common international approaches will be needed to avoid a fragmented regulatory landscape, similar to the divergent national interpretations that have emerged in other areas of technology and data protection.
This global dimension also raises questions of sovereignty. Organisations need to understand where their data is stored, who can access it and which mechanisms, including encryption and key management, can help preserve control over sensitive information.
Responsibility will be shared across the AI supply chain. Developers, technology providers, deployers, legal organisations and individual users will each have a role to play in ensuring that AI is used safely and responsibly.
Building AI Fluency Through Experience and Access
The event concluded with a discussion about AI fluency and education.
The panellists agreed that AI fluency cannot be created solely through written guidance or occasional training sessions. It requires hands-on experimentation, practical use cases and an understanding of how AI can support specific professional tasks.
They also emphasised the importance of accessible education. AI training should not become an elitist resource available only to large organisations or highly specialised professionals. Free online resources, employer-supported programmes and learning through experimentation can help broaden access.
Asked what they would change with a “magic wand”, the panellists offered complementary answers. Their wishes included removing fear around experimentation, recognising that not every task should be automated, creating concrete critical-thinking policies for AI use, closing the global access gap and encouraging legal professionals to identify what they want AI to do less of and what they want to continue doing themselves.
Taken together, these responses captured the central message of the event. The future of legal AI will not be determined by technology alone. It will depend on whether legal organisations can combine better tools with reliable data, strong governance, critical thinking, human expertise and meaningful access to education.
The challenge is therefore not whether AI belongs in the legal ecosystem. It is how the legal profession can build the standards, structures and skills needed to use it responsibly.
About Jus Mundi
Founded in 2019 and recognized as a mission-led company, Jus Mundi is a pioneer in the legal technology industry dedicated to powering global justice through artificial intelligence. Headquartered in Paris, with additional offices in New York, London, and Singapore. Jus Mundi serves over 150,000 users from law firms, multinational corporations, governmental bodies, and academic institutions in more than 90 countries. Through its proprietary AI technology, Jus Mundi provides global legal intelligence, data-driven arbitration professional selection, and business development services.
