The AI Ground is Moving: From Managing Evidence to Supporting Legal Judgement

Kenneth Tombs with some personal observations on the rapid advances in legal AI practice and thinking

Intellectually, the pace of emerging AI technology is unlike anything I have experienced during more than forty years working through successive generations of emerging information technologies. Rarely has commercial capability and academic research advanced so rapidly, or in such close alignment.

Only a matter of months ago, much of the profession’s attention centred on familiar questions. Could AI securely manage documents? Could it search large collections of evidence? Could it extract relevant clauses, summarise witness statements and reduce the burden of disclosure?  What was the truth not circumstance?

Those questions have not disappeared. Instead, they are becoming buried beneath an unprecedented volume of research papers, benchmark reports, and commercial announcements. For many practices, keeping pace with the technology is rapidly becoming as challenging as adopting it. Yet these remain the fundamental capabilities upon which the next generation of legal AI will be built.

June alone saw an exceptional number of high-quality research publications, benchmark studies, and commercial announcements. Taken together, they suggest that legal AI is rapidly moving beyond document capture and management towards supporting the interrogation of evidence and, increasingly, the formation of professional judgement.

What has changed is where both commercial investment and academic research are now concentrating their efforts at pace.

Surveying recent work from Stanford RegLab, Harvey’s Legal Agent Benchmark, long-context AI research and commercial evaluations, a remarkably consistent direction of travel is emerging. Current research is moving beyond document retrieval towards evidence interrogation, relationship discovery, grounded reasoning and, increasingly, support for professional judgement.

That pace of development is remarkable. Having worked through several generations of information technology, I cannot recall another period where commercial viability and academic research have evolved so quickly or with such mutual reinforcement. The cycle between publication, implementation and commercial deployment has become extraordinarily short.

Today, more than twenty-five major AI platforms have the potential to undertake substantial legal tasks, while specialist legal AI providers continue to emerge as next generation support services.  At face value, many are already capable of performing work that, only recently, would have been regarded exclusively as specialist legal support.

Ironically, the problem for many firms may no longer be a lack of AI capability, but an abundance of choice. Which of these apparently capable providers should we trust, trial, and integrate into professional practice?

There is an encouraging international dimension. As George Bernard Shaw famously observed, Britain and America are ‘two nations divided by a common language’. AI may be producing a modern legal equivalent. While the United States has led much of the frontier model development, Europe is increasingly differentiating itself through sovereignty, evidential security, privacy, and governance. For legal practice, those characteristics may prove every bit as important as benchmark performance.

While many of today’s frontier AI models originate in the United States, Europe has developed a strong and increasingly competitive AI ecosystem of its own. European providers have generally placed greater emphasis on data sovereignty, evidential security, privacy, and regulatory assurance,qualities that are likely to resonate strongly with legal professionals. For practices handling sensitive client material, those characteristics will ultimately prove every bit as important as benchmark performance.

The question is no longer whether AI can assist legal work. The challenge now is: How can firms ensure their evidence is AI-ready, and their judgement AI-defensible?

The more interesting question has become something else entirely. How should legal practice adapt now that these capabilities exist?

The significance of this shift is not simply that AI has become more capable. It is that capability is rapidly becoming commoditised. More than twenty-five mainstream AI platforms, alongside a growing number of specialist legal providers, now offer increasingly sophisticated legal capabilities. The differentiator is beginning to move from the technology itself towards how legal practices organise evidence, integrate AI into professional work and demonstrate professional judgement.

Research is increasingly focused on problems that would have seemed ambitious only a year ago: interrogating vast collections of evidence, identifying hidden relationships across thousands of documents, maintaining traceable citations, detecting contradictions, and supporting sophisticated legal analysis. The frontier is no longer simply about retrieving documents; it is about helping professionals understand what those documents collectively mean.

In other words, many of the technically difficult problems are rapidly becoming engineering problems rather than research aspirations.

That shifts the bottleneck away from technology and towards professional practice, governance, and policy. Increasingly, the limiting questions are no longer “Can AI do this?” but “Should it?”, “Can its conclusions be defended?” and “How should evidence be organised so that AI-assisted work remains professionally accountable?”

The challenge is becoming less about making AI capable and more about making legal work ready for capable AI.

Why June 2026 Matters Recent research and commercial developments suggest that legal AI is moving beyond document management towards evidence interrogation, analysis and judgement support. Among the developments are: Harvard Law School / Stanford RegLab: Evaluating Legal AI in Practice Recent work has increasingly focused on measuring how reliably AI supports legal work rather than simply whether it can produce answers. Studies examining citation fidelity, legal reasoning and benchmark performance have highlighted that trust now depends on evidence grounding and verifiable reasoning, not simply fluent text generation. Stanford RegLab: The Citation Gap One of the most widely discussed findings has been the demonstration that even leading legal AI systems can generate incorrect or unsupported legal citations. Rather than slowing adoption, this research has accelerated work on grounded citations, provenance and evidential traceability, reinforcing the importance of professional oversight. Harvey: Legal Agent Benchmark (LAB) Harvey’s publication of the Legal Agent Benchmark marks a notable shift in commercial evaluation. Rather than testing simple question answering, LAB measures how AI systems interrogate extensive legal datasets, cross-reference multiple documents, identify inconsistencies and support realistic legal workflows. The benchmark reflects the industry’s movement towards practical legal reasoning rather than document retrieval. Long-Context AI Research (Google, Anthropic, Meta and others) Recent advances in long-context models mean that AI systems can increasingly analyse entire disclosure bundles, transaction data rooms or extensive evidence collections as coherent datasets. This represents a significant departure from earlier approaches that relied heavily on document chunking and retrieval pipelines. Commercial AI Platforms Continue to Mature Alongside specialist legal providers, more than twenty-five mainstream AI platforms now offer capabilities that are directly relevant to legal practice, including document analysis, evidence interrogation, drafting, multimodal reasoning and increasingly sophisticated analytical support. The availability of these capabilities is becoming widespread rather than exceptional. The Emerging Research Consensus Taken together, these developments suggest a common direction of travel. Research effort is concentrating less on finding documents and increasingly on helping professionals understand evidence, evaluate competing interpretations and support defensible judgement. What this means for legal practice If this trajectory continues, the principal challenge for many firms may no longer be selecting an AI platform. Instead, it will be ensuring that evidence is sufficiently well prepared, connected and governed for increasingly capable AI, and ultimately legal professionals, to analyse it effectively and defend the conclusions reached. “The significance of this shift is not simply that AI has become more capable. It is that capability is rapidly becoming commoditised. The differentiator is beginning to move from the technology itself towards how legal practices organise evidence, integrate AI into professional work and demonstrate professional judgement.”

A Different Competitive Landscape?

That change may also have important consequences for the structure of the legal profession.

For decades, the largest and most complex legal matters have naturally favoured the largest firms. Success depended not only on legal expertise, but on the ability to deploy teams of paralegals, trainees and support staff to organise, review and interrogate enormous volumes of evidence. The economics of complex litigation often reflected the economics of managing information.

That model may now be beginning to change.

Historically, large firms enjoyed structural advantages because complex matters demanded large numbers of people to organise, review and interrogate evidence. As AI increasingly undertakes much of this mechanical work, one of the traditional economic advantages of scale begins to weaken. The value shifts away from processing information and towards interpreting it.

As AI becomes increasingly capable of searching, cross-referencing and analysing extensive evidence collections, some of the advantages previously gained through scale alone begin to diminish. The competitive question becomes less “How many people can we assign to the case?” and increasingly “How well have we organised the evidence?”

This does not diminish the importance of legal expertise. Quite the opposite. As AI assumes more of the mechanical work of evidence interrogation, professional judgement becomes an even more valuable professional skill.

In many respects, the profession may be moving from an economy of labour towards an economy of judgement. Competitive advantage increasingly comes not from deploying more people to review evidence, but from applying better professional judgement to evidence that increasingly capable AI has already helped organise and interrogate.

For smaller and specialist practices, this presents an intriguing possibility. Increasingly capable AI is becoming available as a service rather than something that must be built in-house. Sophisticated analytical capability is therefore becoming accessible to practices of every size.

This does not suggest that smaller firms will replace larger firms. Large practices will continue to possess deep expertise, specialist teams and resources that remain difficult to replicate. However, it does suggest that smaller, well-organised practices may increasingly be able to compete for work that previously lay beyond their practical reach.

AI capability itself is becoming progressively democratised. Professional judgement remains anything but.

This observation connects closely with two recent contributions to the discussion.

In his recent SCL article, Gareth Davies argues that the real professional value lies not in AI’s initial answer but in the lawyer’s correction, interpretation and judgement. Professional expertise remains rooted in understanding where AI is right, where it is wrong and, perhaps most importantly, why.

Similarly, Tara Chittenden, Foresight Lead at the Law Society has observed in conversation with the author that while legal practice possesses mature frameworks for gathering, testing and presenting evidence, it has comparatively little operational guidance describing how evidence is systematically transformed into professional judgement.

Taken together, Gareth Davies explains where professional judgement resides, while Tara Chittenden highlights the absence of a clear operational framework describing how evidence becomes judgement. Those observations reinforce precisely the direction in which current AI research now appears to be moving.

Evidence management remains essential.

Increasingly, however, the challenge concerns how evidence is organised, connected and interpreted before professional judgement is formed.

Historically, considerable effort has often been invested in manipulating documents into formats that software could process effectively. As AI systems become better at reading unstructured information directly, that effort appears likely to reduce significantly. Instead, greater emphasis is likely to fall on ensuring that evidence is captured coherently at source, that provenance is preserved and that relationships between documents, events and decisions remain intact throughout the life of a matter.

The quality of legal AI may therefore become increasingly dependent on the quality of evidence preparation rather than on the sophistication of the AI itself.

One practical example of this thinking is the emerging Dossier Space concept. Rather than treating AI as another search tool, it explores how evidence, chronology, provenance and professional judgement might exist within a single evidential workspace where lawyers and AI systems contribute together to analysis. The objective is not to replace professional judgement but to create an environment in which increasingly capable AI can operate against well-prepared, well-connected and professionally governed evidence.

The emphasis therefore shifts from manipulating data for machines towards preparing evidence for professional reasoning.

That feels like an important distinction but there is perhaps an irony here.

After two years of asking whether AI could understand legal evidence, we may now be approaching a different question entirely. How should legal practice organise itself when AI already understands far more than we expected?

The next competitive advantage may not come from purchasing another AI platform. More than twenty-five capable systems already exist, and that number continues to grow.

It may instead come from something much more familiar to lawyers: the disciplined organisation of evidence, the transparent exercise of professional judgement and the ability to demonstrate how one became the other.

If that proves correct, the coming years may be remembered not as the period when AI replaced lawyers, but as the period when increasingly capable AI reduced one of the historic advantages of scale, enabling well-organised practices – large and small – to compete more effectively on the quality of their professional judgement rather than simply the size of their technology investment.

Further Reading

• Stanford RegLab: The Citation Gap: Evaluating the Reliability of Legal AI Citations (2025)
• Harvey. Legal Agent Benchmark (LAB) (2026)
• Google DeepMind. Gemini 2.5 Technical Report (2026)
• Anthropic. Claude 4 System Card (2026)
• Gareth Davies. The The Virtuous Circle: Law, AI, and the Apprenticeship of Judgment. Society for Computers & Law (2026).