At a recent legal forum in Hong Kong, a simple show of hands told a complex story. When asked how many attendees had used an AI tool in their legal work in the last 30 days, nearly every hand in the room went up. When asked how many had a clear organisational policy governing that use, far fewer did.
This gap, between adoption and strategy and governance is the defining challenge facing the legal profession this year.
The era of experimentation is over
The use of GenAI in legal work has doubled in the last 12 months, 56% of legal professionals are using publicly available AI tools, and, perhaps more tellingly, 54% of clients want their law firms to be using GenAI. The demand no longer sits with the individual professional or worker, it’s coming from the market itself.
What’s changed is not just the pace of adoption, but the nature of the work being delegated to AI. Workers are no longer using it to just handle peripheral tasks, it is being embedded into core legal work: research, document review, summarisation, and drafting. The tools are inside the standard workflow of a legal professional.
As GenAI evolves into autonomous, agentic workflows and deep research capabilities redefine legal analysis, our profession stands at a pivotal moment: AI is no longer a tool, it is becoming legal operations infrastructure. Currently, only 15% of legal professionals are using agentic AI, but 53% are planning to use it in the next 12 to 24 months.
The divide is already here
Thomson Reuters Global CEO Steve Hasker has described 2026 as the year of a new divide between organisations that have adopted a clear AI strategy and those that have not. And the data behind that statement is striking: organisations with clear AI strategies are twice as likely to experience revenue growth and 3.5 times more likely to realise critical AI benefits. Yet, according to our research, only 22% of organisations have achieved that strategic clarity.
This means that the majority of firms and legal departments are already falling behind because they haven’t made a clear decision about it.
Across Asia, the barrier is rarely access to the technology itself. It is culture. It is fear of choosing the wrong tool. It is change management. However, waiting for certainty is itself a choice, and that also carries a cost. The competitive divide is here, and waiting will only set you further back.
The right question to ask
In Hong Kong, the regulatory environment is actively evolving:
• The Office of the Privacy Commissioner for Personal Data has issued frameworks and checklists.
• The Hong Kong judiciary is currently drafting guidelines.
• The Department of Justice is reviewing whether existing law is sufficient.
Hong Kong’s governing bodies are actively exploring AI with a focus on governance frameworks and pilot projects that could extend to the court system. Which means that today, if something goes wrong, a hallucinated case citation filed in court, a flawed contract review, the professional liability lands on the practitioner.
This shifts the conversation from “should we use AI?” to a far more important question: “can I defend the results of the AI that I’m using?” That question comes down to four factors:
• Where does the AI get its information?
• Can every output be traced to a verified source?
• Is client data protected?
• And is there a human in the loop at every critical decision point?
These four questions are not just a governance framework. They are the standard by which AI tools for legal professionals should be evaluated.
What fiduciary-grade AI looks like in practice
Not all AI is built the same, it’s useful to understand how different AI models are built, and there are three main frameworks, but we do things differently.
- Data Sources and Grounding
General-purpose AI tools like ChatGPT or Gemini are trained on broad internet data, which is vast, but unverified. They can provide hallucinations by confidently cite a case that doesn’t exist.
By contrast, with CoCounsel Legal the architecture is fundamentally different: one guesses at what the law might say, the other is grounded in what the law actually says[CM1.1], trained by practical expertise in how it is applied. - Agentic Design and Human Oversight
Some AI tools are built to give you a fast answer, one prompt, one response and no transparency into how it got there. Deep Research, by contrast, is built as an agentic system that generates a multi-step research plan that has been trained by legal professionals before it runs, giving the practitioner the visibility in how the research is conducted, demonstrating its refinements to the process for optimal results, but finalising the research and offering a full cited report. - Data Privacy & Model Training
Many consumer-grade AI tools use your inputs to improve their models, meaning your client’s confidential information could, in some configurations, become part of a training dataset. Thomson Reuters is built on the principle that customer data is not used to train third-party models.
In the legal profession, “almost right” is not good enough. Adoption of legal AI continues to accelerate, alongside a widening gap between the speed and convenience of general-purpose AI and the accuracy and verifiability of professional-grade systems.
CoCounsel Legal is designed to meet the standards of legal professionals, where almost right is not good enough.
Built on trust, expanded for what comes next
Thomson Reuters offers capabilities that have taken 175 years to build: authoritative professional content across Westlaw, Practical Law, and KeyCite, curated and validated by practicing attorneys and legal specialists. More than 2,600 experts globally shape how CoCounsel reasons, ensuring outputs reflect the standards of real professional work. Customer data is not used to train third-party models and is not shared beyond a customer’s own environment and in professions where confidentiality is both a legal and ethical obligation, that commitment is a baseline requirement.
This week, Thomson Reuters took another significant step forward. Thomson Reuters announced a new Model Context Protocol (MCP) integration with Anthropic that connects Claude directly to CoCounsel Legal. Legal professionals can now move seamlessly between general-purpose AI and citation-grounded legal work, from either working environment. As CTO Joel Hron put it, “Legal professionals deserve AI they can trust with their most important work. In professional environments, trust in AI is a property of the system itself, built into the architecture and verifiable at every step.”
Today, one million professionals across 107 countries and territories use CoCounsel, Thomson Reuters AI technology. This demonstration of the scale of trust is not accidental; it is the result of purposefully building AI that meets the standard the profession demands.
The decision is yours
The firms that will define the next decade of legal practice in Hong Kong are taking action now, not waiting for certainty.
The gap between adoption and governance is where risk lives. It is also where competitive advantage is won.
