In his speech as part of the 2025 LawTechUK event, Sir Geoffrey Vos made a striking observation; when he spoke about the risks of AI, lawyers nodded vigorously, however when he suggested AI could be used to expedite legal advice and reduce the cost of dispute resolution, the room froze.
This reaction captures the paradox I see across the legal sector today.
AI has the potential to free up fee earners’ time, improve client experience, and support better decision-making across matters and litigation strategy. Yet for all the discussion, genuinely embedded AI adoption remains rare. This isn’t because legal leaders fail to see the opportunity, it’s because applying AI in a profession built on judgement, accountability, and trust is uniquely complex.
Having spent years working with national legal firms, I see the same patterns emerging again and again in conversations with partners, CIOs, and innovation leads. The five barriers outlined below frequently arise at pivotal stages of AI programmes and are among the primary factors causing initiatives to stall, budgets to come under scrutiny, or risk concerns to overshadow progress.
If parts of this feel familiar, that is intentional. These patterns are useful indicators of where clarity, confidence, or governance may be missing, and where focused attention could have the greatest impact.
This article is designed to set the context, the articles that follow in this series will focus on suggested approaches firms are taking to address each barrier in practice, with an emphasis on governance, decision making, and low risk ways to move from experimentation to meaningful adoption.
1. The Expertise Gap: Understanding What AI Actually Does
The challenge at the heart of AI adoption in legal is not a lack of legal expertise, it’s the emergence of a new class of technology that demands unfamiliar skills, governance models, and evaluation frameworks.
According to a 2025 Forbes survey, 32% of law firms acknowledge they lack the expertise needed to advance AI adoption. This isn’t an indictment of capability, it reflects how quickly AI has evolved beyond traditional legal technology.
Legal expertise has always involved complex reasoning, nuanced judgement, and deep precedent analysis. AI systems can now reason through multi-step problems, plan workflows, and execute tasks that previously required human intervention. This creates a steep learning curve, particularly for leaders who remain accountable for decisions made with or by AI systems.
In practice, successful adoption starts with a clear view of what AI can and cannot be relied upon to do in a legal context. When leaders are balancing ambition with risk, expectations can move faster than the technology’s current maturity. This tension often surfaces in practical questions around governance, accountability, and where AI should, and should not, be used. Over time, this gap can leave AI confined to pilots rather than embedded into day-to-day legal operations.
A useful signal: This barrier often shows up when conversations about AI remain abstract, responsibility feels unclear, or decisions stall because there is no shared confidence around where AI should and should not be applied.
2. The Strategy Vacuum: Moving Beyond Experimentation
Firms with a clear AI strategy are twice as likely to achieve revenue growth and three times more likely to realise measurable benefits from AI, according to Thomson Reuters’ 2025 Future of Professionals report. Yet only 22% of firms report having reached that level of clarity.
Developing a robust AI strategy is difficult while organisational understanding of AI is still evolving across partners, IT, risk, and compliance functions. Without that shared understanding, it’s hard to make confident long-term decisions about use cases, procurement, governance, or talent.
As a result, many firms take a cautious and entirely understandable approach: piloting tools with small groups of users to limit risk. The problem is that these pilots are often disconnected from a broader strategy. They generate interesting signals, but not enough evidence to assess firm-wide value, ROI, or long-term impact. Adoption remains surface-level, not because the technology lacks promise, but because the firm lacks a framework for scaling it successfully.
A useful signal: When AI activity is spread across pilots, tools, or teams without a clear link to firm level priorities, it is often a sign that experimentation has outpaced strategic alignment.
3. The Data Dilemma: Trust, Control, and Readiness
Another major barrier to AI adoption is data. The legal profession is built on confidentiality, privilege, and regulatory responsibility, so caution around how client data is handled is not resistance to innovation, it is professional judgement.
In a 2025 Embroker survey, 41% of US firms cited data privacy concerns as a key barrier to AI adoption. Even industry-specific tools such as Harvey and Luminance raise questions about data ownership, control, and risk exposure.
Beyond trust, there is also a readiness issue. Many law firms are not structured for quantitative analysis, data is often spread across multiple systems, reported manually, and tracked using inconsistent metrics. Forbes research suggests many firms operate across five to ten disconnected applications, with only a small fraction achieving seamless integration.
As Forbes highlights, AI is not a remedy for systemic disorder, it amplifies whatever processes already exist, for better or worse. Without reliable, well-governed data, AI outputs will always be limited. Equally, without strong and dependable underlying infrastructure to support the fast, data-heavy demands of AI, the inevitable inconsistent results also undermine trust in AI itself.
A useful signal: Data concerns tend to surface when progress slows under questions of ownership, accuracy, or risk, particularly when AI outputs are difficult to explain or defend internally.
4. Accuracy, Hallucinations, and Accountability
Concerns around accuracy and hallucinations are not hypothetical, there have been high-profile cases of lawyers being fined for submitting AI-generated briefs containing fictitious citations. The result is what many firms now describe as an “AI tax”: the added burden of auditing AI-assisted work alongside traditional review processes.
These cases are best understood as governance failures, not as a verdict on AI itself. AI tools are not inherently unreliable, the real issue is how they are used, what data they are trained on, and what oversight frameworks surround them.
AI will only ever be as reliable as the data, controls, and accountability structures wrapped around it. In a profession with a low tolerance for error, these structures matter as much as the technology itself.
A useful signal: When AI generated work requires additional layers of review or creates uncertainty around accountability, the issue is rarely the technology alone but the controls and expectations surrounding its use.
5. Culture and Confidence: From Caution to Capability
Finally, there is a cultural dimension. Legal leaders are trained to manage risk, protect clients, and uphold professional standards. This mindset is a strength, but it can also slow adoption when AI is framed as something that threatens these values rather than supporting them.
What I increasingly see is not resistance, but uncertainty. Leaders are open to AI, but cautious about moving too quickly without clarity on responsibility, impact, and consequences. Bridging that gap requires not evangelism, but practical confidence built through understanding, governance, and controlled experimentation.
A useful signal: What appears as cultural resistance often shows up in practice as hesitation. A reluctance to move faster without clearer guardrails, ownership, and consequences.
In Conclusion
The legal sector’s experience with AI is not a rejection of innovation, it reflects the reality of applying a fast-moving technology within a profession defined by trust, judgement, and accountability.
The firms best positioned to unlock AI’s value will be those that invest in understanding before acceleration, strategy before scale, and data foundations before automation. These challenges are not insurmountable, but they do require deliberate leadership decisions. Decisions we will discuss in the next article, coming very soon.
In This Series
- Part 1 (this post): The five barriers blocking legal AI adoption
- Part 2: Practical strategies for overcoming these challenges
- Part 3: High impact AI use cases in Legal
