AI enters one of asset management’s most complex fields. Reviewing a Ucits prospectus has long been one of the most demanding legal exercises in Luxembourg’s fund industry. Each document must comply with thousands of pages of European and Luxembourg regulation while remaining consistent across hundreds of pages and multiple sub-funds.
As regulation continues to expand, fund managers face growing pressure to deliver compliant documentation faster and at lower cost, without compromising legal certainty. According to Arendt, this growing complexity is precisely what motivated the development of Ava (Arendt Visa Agent a reference to the now-discontinued CSSF “Visa” process), an AI-powered compliance review platform designed specifically for Ucits prospectuses.
“Ucits prospectuses operate in one of the most demanding regulatory environments in financial services. Our objective was not to replace legal judgement but to help clients manage growing regulatory complexity while maintaining the highest standards of consistency and quality,” says 
Turning legal expertise into machine-readable controls
The innovation does not primarily lie in the artificial intelligence itself, but in the way decades of legal expertise have been translated into a structured review methodology. Arendt identified the relevant Ucits disclosure obligation contained across more than 5,000 pages of legislation before converting them into approximately 1,500 “atomic controls” – binary compliance checks that AI can analyse individually.
“Large language models are extremely powerful, but they are not designed to process an entire Ucits prospectus in a reliable way. We therefore broke down both sides of the equation: the regulatory framework into approximately 1,500 atomic controls; and the prospectus itself, into the precise provisions that each control needs to test, sub-fund by sub-fund. That structure is what makes the analysis both stable and auditable,” explains 
This structured approach allows the system to analyse highly complex legal requirements in a far more consistent way than a general-purpose AI model.
Why one AI model wasn’t enough?
Interestingly, the project initially relied on a single large language model. According to Arendt, that approach quickly proved insufficient.
The firm instead developed a multi-agent architecture in which different AI agents perform specialised tasks. One identifies the various sub-funds within a prospectus. Another isolates only the relevant sections, reducing the review from roughly 500 pages to around 20 pages before the legal analysis even begins. Additional agents repeat the analysis several times to improve stability, audit the conclusions against the firm’s internal standards and provide additional legal context where necessary.
“Our first tests with a single language model simply weren’t consistent enough. By assigning specialised tasks to different AI agents – from identifying sub-funds to auditing the final result – we significantly improved both stability and accuracy,” says Xavier Mangeard.
AI as an assistant, not a decision-maker
Despite the technological sophistication, Arendt insists that lawyers remain firmly at the centre of the process.
The firm’s legal expertise forms the foundation of the system, while senior lawyers continue to define the review, validate the conclusions and interpret complex regulatory questions. More than a year of testing and continuous governance by experienced lawyers are intended to ensure that the platform remains aligned with evolving regulatory expectations. The technology is therefore designed to augment legal professionals rather than replace them.
A glimpse of how regulation may evolve
Beyond the technology itself, AVA also illustrates a broader shift taking place across professional services. By enabling lawyers to structure decades of accumulated knowledge into digital assets that can be deployed repeatedly and at scale. AI will allow them to focus more of their time on higher-value work: applying judgement, advising clients and solving complex legal questions.
Arendt believes regulators themselves will increasingly rely on artificial intelligence.
With growing volumes of documentation and an accelerating pace of European regulation, manual supervision alone may no longer be sustainable. The firm argues that AI-assisted tools could eventually become essential for supervisory authorities seeking to review an ever-growing number of documents while maintaining consistency.
Luxembourg’s competitive advantage
The firm also believes Luxembourg’s concentration of fund administration, legal and regulatory expertise gives it a unique opportunity to become a leader in legal AI for financial regulation. While technology companies possess strong AI capabilities, they often lack the legal knowledge built through years of advising clients on complex Ucits structures.
The law is public: anyone can read it. What is not public is how to break it down into rules that a machine can test, and how those requirements are actually applied in practice in Luxembourg. That second layer is not written down anywhere. It is acquired through experience.
According to Arendt, that accumulated expertise – not the technology itself – is what creates the competitive advantage.
Lawyers won’t disappear but their work will change
Rather than eliminating legal professionals, AI is expected to shift their role. Routine document production is likely to become increasingly automated, while lawyers will spend more time validating results, interpreting complex situations and exercising professional judgement.
AI can identify issues, but it cannot assess their relative importance from a practical perspective. Knowing what is incorrect has become accessible; knowing what is important remains the lawyer’s role.
“The lawyer remains fully responsible for the final assessment. AI accelerates the review and strengthens consistency, but professional judgement remains essential. As regulate continues to grow, the real value will lie in combining technology with deep legal expertise,” concludes Corinne Prinz.
Looking ahead, Arendt believes that, by the end of the decade, AI will handle much of the basic groundwork involved in prospectus reviews. Lawyers, meanwhile, will increasingly focus on what machines still cannot do: applying judgement, challenging conclusions, providing innovative solutions to complex questions and assuming ultimate legal responsibility.
