The stakes are real. A missed disclosure isn’t just an inconvenience, it can surface as a inspection finding, a peer review deficiency, or a restatement. Firms know this, which is exactly why so many roll forward last year’s checklist rather than start fresh, or burn valuable senior and manager time doing it all over again. Neither option solves the underlying problem: this is manual, judgment-heavy work at a scale that doesn’t favor manual review.

Bringing AI into the checklist itself

At , today we’re launching Disclosure Review, a new AI capability available to all customers with the Disclosure module, at no additional cost.

Disclosure Review matches disclosure requirements against what’s actually present in the financial statements, marking each as identified or not found, and pairing every recommendation with plain-language rationale and citations back to the source. That last part matters as much as the matching itself. Disclosure requirements are often written in dense, technical standards language. When junior staff can see why something was marked a certain way, in accessible terms, they build real understanding of the requirement, not just trust in the output.

Where the AI and the auditor’s own judgment diverge, the discrepancy is flagged automatically. This means teams can focus their attention exactly where it’s needed instead of re-checking everything from scratch.

We’ve spent months validating this in an extended customer beta, and the results have been encouraging. Roughly 30% of eligible Disclosure module customers are already using it, generating close to 300 disclosure review analyses a month across mid-sized and large firms alike.

Meeting the market, with room to grow

Future capabilities will extend further into the harder judgment calls, like “no” and “not applicable” determinations, where auditor expertise matters most.

Reviewing disclosure requirements will likely never be anyone’s favorite part of the engagement. But it doesn’t have to be the part that consumes the most time or carries the most risk. That’s the shift we’re focused on – freeing up capacity so audit teams can spend more of their time on judgment and insight, not repetitive matching.

This post was authored by Corey Wells, General Manager of Audit at ��������.

In June, I wrote about our strategy to make Fiduciary-Grade AI available wherever professional work begins, and today we’re taking another step with OpenAI.����

In partnership with OpenAI, �������� is showing how both CoCounsel Legal and HighQ can connect with ChatGPT Enterprise. We are previewing a coming CoCounsel Legal experience, which demonstrates how legal professionals will access CoCounsel Legal capabilities directly from ChatGPT, and starting today HighQ is available through a Model Context Protocol (MCP) connection.��

This collaboration is a clear reinforcement of our strategy to build an extensible, open, and connected ecosystem that delivers Fiduciary-Grade AI wherever our customers work.����

“We believe the next generation of AI for law will be built around what makes each firm distinctive – its expertise, judgment, and way of working. By building alongside ��������, we’re turning that expertise into AI lawyers can use every day, creating new ways for firms to serve their clients and shape the future of legal work,” said Jason Boehmig, GM Legal Industry at OpenAI .

From a Connection to an Architecture��

Together, the experiences we are developing with OpenAI point towards a future where legal professionals working in ChatGPT Enterprise could connect to the matter context they are authorized to access in HighQ and call on CoCounsel Legal for specialized legal capabilities. The systems remain distinct, but the experience becomes more connected regardless of the surface of interaction.��

That is the architecture we are working toward: frontier intelligence connected with the context and professional-grade capabilities needed to complete high-stakes work.��

MCP Is the Infrastructure, Not the Story��

Open standards like MCP are important because they create a more consistent way for AI systems to discover and invoke tools and context across platforms. But the protocol itself is not the end goal. What matters is what it allows professionals to do.��

Lawyers at global firms and in-house counsel should be able to begin work in the environment that is most natural for the task, bring in the context they are permitted to use, call on trusted professional capabilities, and move the work forward, all while maintaining the governance and security controls their organizations require.��

That experience must be built around security, permissions, governance, transparency, and accountability. Open cannot mean uncontrolled. Interoperability is valuable only when trust travels with the work.��

CoCounsel Legal remains the trusted professional AI system designed to ground, validate, and help complete high-stakes work. Our ecosystem strategy makes that system more accessible, while preserving the role it plays at the center of professional work.��

A Strategy Built for Choice��

Our customers will use different models and AI environments for different kinds of work. Our vision and strategy is to ensure Fiduciary-Grade AI travels with the work, regardless of which environment our customers choose. Environments will continue to evolve as frontier intelligence advances. What should persist is the ability to connect them with the trusted content, customer context, workflows, and professional capabilities our customers rely on.��

The future of legal AI will not be defined by one model, one interface, or one company working alone. It will be defined by how well an ecosystem of general-purpose and specialized systems can work together on behalf of the professional.��

The opportunity is to give customers the best of both: the freedom to work in the environments they choose and access to systems that offer accuracy and accountability when it matters most.��

What Comes Next��

The experiences we are showing with OpenAI are an important step, but they are not the destination.��

We will continue to work with our customers to provide them with a best-in-class destination for legal work – CoCounsel Legal, extending its capabilities across the models, platforms, and workflows our customers choose to use.��

Our plugin with ChatGPT Enterprise offers a glimpse of what that connected future can look like: matter context and trusted legal intelligence available from the same AI environment, with CoCounsel Legal at the center of how high-stakes work gets completed.��

Agentic AI Grounded in Deep Legal Expertise

The next generation of CoCounsel Legal (US)

The next generation of CoCounsel Legal has launched in the U.S. A complete rebuild of CoCounsel Legal, it’s engineered to work at the level of a senior associate, reasoning through legal issues the way an attorney does and grounded in authoritative Westlaw primary law, trusted Practical Law guidance, and your firm’s own knowledge.��Describe a matter in plain language, whether��it’s��contract review, litigation strategy, deal structuring, or sifting thousands of discovery documents, and CoCounsel Legal handles the rest.��It develops the approach, works through each step, and links every citation so you can verify it yourself.����You can give each matter its own workspace, keeping documents, templates, and conversations together. ����It also meets you in the tools you already use, including Microsoft 365 and Claude. Your data always��stays��your firm’s — private, and never used to train models.��This is Fiduciary-Grade AI your firm can stake its reputation on.

Learn more at

Westlaw Brief Builder (US)

A transformation of research and drafting, Westlaw Brief Builder helps litigators move from research and issue analysis to first-draft brief creation while validating authority along the way. It follows a workflow purpose-built for briefs, guiding the drafter through each stage of the process in a way that litigators actually think and work — an organized, connected experience developed and validated by attorneys. Powered by Westlaw Deep Research, KeyCite, and Practical Law — authority no other legal AI can match — it proposes relevant facts, arguments, and supporting authority while keeping lawyers firmly in control of strategy, legal theory, and final decisions. And because output is verified against the law and the facts, then surfaced for review and confirmation, Westlaw Brief Builder delivers briefs not only faster, but briefs you can stand behind.

Global Connected Platform

Firm Document Management Connectors via Syncly (US, UK & Canada)

Admins can now connect CoCounsel Legal directly to a firm’s document management system through new Syncly connectors, including iManage on-prem, iManage Cloud, and SharePoint. Once connected, CoCounsel can access and act on firm content directly, with a connector icon appearing for end users wherever a connection is active — simplifying the ability to bring trusted firm documents into AI-assisted research, analysis, and drafting without leaving the platform.

Expanded CoCounsel Legal MCP with Claude (US)

CoCounsel Legal’s MCP connection with Claude is expanding. What started with Deep Research now extends to research, analysis, and drafting, bringing more of CoCounsel Legal’s fiduciary-grade AI capabilities directly into Claude. Lawyers can describe a matter in plain language inside Claude, and CoCounsel Legal plans the work — reasoning from authoritative Westlaw primary law, trusted Practical Law guidance, and a firm’s own knowledge — and returns cited, traceable work product without leaving Claude.Tabular Analysis Enhancements in CoCounsel Legal (US)

Tabular Analysis is evolving. In the newly launched version of CoCounsel Legal, bulk document review is now built directly into the workspace — no more exporting files or rebuilding a separate database to review a subset. Creating and editing tables is simpler too, with a redesigned experience built from direct customer feedback, and multiple tables can now share the same set of files instead of duplicating work for each one. Tabular Analysis also runs by default on Thomson, �������� own purpose-built legal AI model — the organizational and technical foundation for a more connected CoCounsel Legal.

Built for How You Work

Verbatim Extraction in Tabular Analysis (US, UK & Canada)

Attorneys building disclosure schedules, privilege logs, and red-flag memos need exact clause language, not paraphrases. A new Verbatim column type in Tabular Analysis returns exact source text instead of AI-generated summaries, so attorneys no longer need to reopen the original document to re-quote a clause. Users select “Verbatim” when configuring a column and describe what to pull; CoCounsel returns the exact language from the source, validated against the document, and clicking any cell opens the source with the precise span highlighted for verification. Exports carry the verbatim text into Word and Excel with quotation marks already applied, ready for partner review and closing binders.

Judicial Experience on CoCounsel (US)

A new, personalized CoCounsel environment is now available for trial, appellate, and supreme courts. When judges or clerks activate the Judicial setting, the application adapts to their role with a neutral-stance default and a suite of curated judicial workflows, including pro se brief standardization, statute of limitations checks, cross-brief issue synthesis, and case fact drafting — every output grounded in trusted Westlaw and Practical Law content, with hyperlinked citations and KeyCite verification for auditable, defensible decisions.

Thomson – New Proprietary LLM (US, Canada & UK)

Thomson, the first proprietary large language model built by �������� draws on decades of proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters plus hundreds of subject matter experts to reach Fiduciary-Grade standards at a fraction of typical cost. Early evaluations, including external testing by legal and AI academics, put Thomson on par with leading frontier models, particularly in instruction-following and domain-specific legal reasoning, and the model is now being deployed first in Tabular Analysis within CoCounsel Legal, with plans to extend Thomson across the legal and tax portfolio alongside additional sovereign AI options.

Learn about the Thomson LLM .

Deep Research Verify in Practical Law Premium (UK)

Deep Research Verify is now available in Practical Law Premium, integrated directly into the existing Deep Research workflow. Verify lets users review specific AI-generated legal assertions against relevant supporting passages from underlying Westlaw UK and Practical Law sources, with source support displayed inline alongside the original report. By reducing manual cross-referencing and providing a direct route to further research, Verify helps legal professionals assess AI-generated research more efficiently, transparently, and confidently.

Deep Research Verify in CoCounsel (UK)

CoCounsel UK now offers the ability to verify a Deep Research report directly within the platform. Deep Research Verify checks a Deep Research report against Practical Law and Westlaw UK’s authoritative sources, confirming citations are accurate rather than hallucinated and surfacing additional relevant law the report may have missed — moving users through verification faster and giving them more links to explore for further research.

Westlaw Advantage Canada — Parallel Search (Canada)

Parallel Search is a new feature in Westlaw Advantage Canada that finds cases with facts or issues similar to a scenario described in plain language — not just documents matching the words in a query. Users describe a fact pattern or issue in a sentence, and Parallel Search surfaces up to 25 Canadian cases with analogous facts or reasoning, helping researchers find true comparators and surfacing analogous outcomes they might otherwise miss, especially on novel or hard-to-phrase issues.

Deep Research Verify in Westlaw Advantage Canada and CoCounsel Legal (Canada)

Deep Research Verify is now available within the Deep Research experience on Westlaw Advantage Canada and CoCounsel Legal. For each assertion in a Deep Research report, Verify presents supporting text from the cited source and flags where support may be lacking, with a clear path for additional manual verification — helping users review AI-generated research with greater confidence and less manual cross-referencing.

Explore These New CoCounsel Legal Features Today

Sign in to CoCounsel Legal today to put these new research, drafting, and document-management capabilities to work, or explore training options at the .

To learn more, please visit the .

What happens when businesses take a laissez-faire approach to AI? Individual workers fill the gap by using publicly available chatbots that could create serious compliance risks and liability issues.

The phenomenon is known as shadow AI, and according to the �������� Future of Professionals Report 2026, it’s currently occurring among more than one-third (36%) of professionals working in corporate tax, legal, and compliance functions who admit they are actively using AI tools their organization have not sanctioned. For businesses at the center of this issue, the risk of information leakage, inaccuracies, cut corners, and a collapse of standardized processes could create dangerous ripple effects.

Pressure to Move Faster

On the surface, the findings should not come as a huge surprise. AI is everywhere these days, and it’s already become second nature for many of us to turn to widely available consumer chatbots for guidance on everything from dinner recipes to exercise tips. On top of that, many corporate professionals are facing increased pressure from internal stakeholders and clients to deliver faster, better-informed decisions with better efficiency and cost controls. According to our research, 58% of corporate tax and legal professionals say they’re facing “some” or “significant” pressure from their key stakeholders to move faster on AI adoption.

The disconnect occurs when tools designed for consumer-grade tasks are applied to professional-grade work, which often contains proprietary or sensitive information that should not be shared on external servers, or highly specialized data that consumer large language models (LLMs) were never meant to process. Still, despite the obvious risks associated with using unsanctioned AI tools for high-stakes professional work, many companies are just not moving fast enough on AI adoption and, as a result, employees are taking matters into their own hands.

Understanding the Risks

For many, it’s a survival instinct. In fact, 15% of professionals in corporate enabling functions say they are already seeing financial consequences of insufficient progress on AI adoption by their companies, and another 29% say they expect them within 12 months. Pressed to continually find ways to do more with less, inundated with news about new AI tools that can do seemingly anything, and drawn-in by the allure of freely available and amazingly powerful consumer tools, it’s no surprise that many corporate tax, legal and compliance professionals would start experimenting.

The downsides of that trend are already starting to become , and can include everything from lapses in corporate governance to over-reliance on incorrect or incomplete information – not to mention a lack of standardization whereby each individual employee starts using their own tool.

A New Focus on Collaboration

To address these issues, corporate enabling functions must start to make a clear case to business leadership for why they need professional-grade AI solutions. The fact is that as the AI ecosystem matures, solutions developed for professional grade tasks like corporate tax, law, compliance, and others are becoming highly specialized. These are not the mainstream, consumer-grade chatbots; they are finely tuned pieces of professional software developed for highly specific use cases. Senior leadership may have big-picture AI mandate, but they may not necessarily understand the need for specialized tools. Only the teams in the trenches can deliver that perspective, and these teams need to get a seat at the table where they can advocate for themselves.

It’s also high time for most corporate tax, legal and compliance professionals to start taking an honest look at what kinds of tools their teams are currently using – both sanctioned and unsanctioned – to determine where are AI tools already being used, where are people improvising, and where is there unmet demand.

Functions working in isolation on AI strategy are creating shared risk: inconsistent accountability, incompatible governance, and shadow AI that nobody owns. Fiduciary functions like legal, tax, and compliance hold the professional standards that should anchor the enterprise’s AI governance. While, currently, the C-suite, technology and operations teams hold disproportionate sway over AI budgets and implementation, a broader conversation is needed. The general counsel, Chief Compliance Officer and corporate tax leaders are particularly well positioned to lead this conversation – with an emphasis on what’s at stake if companies get it wrong. The time to start having that conversation is now.

About the author
Liz Zimick is President, Corporates, at ��������

For litigators, that matters because some of the most important context in a matter does not begin in a research or drafting tool. It begins in the evidence.��

Documents. Testimony. Facts developed over the course of discovery. The record a lawyer ultimately has to connect to the law and turn into analysis, strategy, and work product they can stand behind. AI should make that process easier.��

This week at ILTACON, �������� and Everlaw announced plans to integrate Everlaw with CoCounsel Legal. The first planned integration will allow mutual customers to bring Everlaw documents into CoCounsel Legal in bulk, making it easier to use litigation and investigation materials across CoCounsel Legal’s research, analysis, drafting, and workflow capabilities.��

For customers, that means less friction between reviewing evidence and doing the legal work that follows.��

For us, it is also one more piece of the broader CoCounsel Legal ecosystem we are building: connecting the systems, content, and tools professionals already rely on so they can move through complex legal work with more continuity and less unnecessary handoff.��

Everlaw brings an important part of that ecosystem into the picture for litigation: the evidentiary record.

Connecting evidence, context, and legal analysis��

A litigator rarely asks a purely abstract legal question. The real question is how the facts, evidence, and applicable law come together: whether a particular document changes the argument, whether testimony is consistent with the rest of the record, or whether the evidence supports the position being put before a court. Carrying that context through the legal workflow is essential to making AI genuinely useful in litigation.��

CoCounsel Legal brings together research, analysis, drafting, and other legal workflows with authoritative content from Westlaw and Practical Law. But in litigation, authoritative legal information is only one part of the picture. The evidentiary record matters just as much.��

Bringing those worlds closer together can make AI much more useful in the day-to-day practice of law.��

By reducing the need to manually move documents between systems, we can help lawyers spend less time on those handoffs and get to the substantive legal work faster.��

Moving from evidence to analysis with less friction��

Consider what happens when a litigation team identifies a set of important documents during discovery.��

Those materials may already have been collected, reviewed, organized, and understood within an eDiscovery platform. But when the team moves into other parts of the legal workflow, that context does not always move with them.��

That can mean exporting documents, uploading them into another environment, and reconstructing parts of the matter before the lawyer can move forward. Every handoff creates friction.��

The planned Everlaw integration is intended to shorten that path. Once Everlaw documents are available in CoCounsel Legal, lawyers can apply CoCounsel Legal’s research, analysis, drafting, and workflow capabilities to those materials, creating a more direct connection between the evidentiary record and the legal work that follows.��

It is also exactly the kind of connection we want to keep adding across the CoCounsel Legal ecosystem: bringing more of the places where legal work already happens into a workflow where professionals can research, analyze, draft, and act with the right context around them.��

Connected AI still has to meet the standard of legal work��

Making more information available to AI cannot mean lowering the standard for what comes out. The lawyer is still responsible for the argument. Still responsible for the citation. Still responsible for understanding whether the evidence supports the conclusion.��

That is why verification, transparency, and professional judgment matter so much. At ��������, we describe this standard as Fiduciary-Grade AI: AI designed for high-stakes professional work, grounded in authoritative information and built so professionals can review, verify, and ultimately stand behind the work it helps produce.��

Connecting more of the matter into that workflow should strengthen the lawyer’s ability to exercise judgment, not remove the lawyer from the process.��

Building a broader ecosystem around legal work��

Law firms and legal departments have invested in specialized technology for a reason.��

Evidence may live in an eDiscovery platform. Documents may live in a document management system. Legal research comes from trusted legal sources. Work product may move through several systems before it is complete. AI is not going to make that ecosystem disappear. The opportunity is to make it work together better.��

That is the direction we are taking with CoCounsel Legal. We are building an ecosystem designed to connect the tools, content, and workflows professionals already depend on, while preserving the context, permissions, controls, and trust required for high-stakes work.��

Everlaw is an important addition to that ecosystem because it brings the evidentiary record closer to the research, analysis, and drafting lawyers are already doing in CoCounsel Legal.��

And it is one part of a much broader direction.��

As we continue expanding the CoCounsel Legal ecosystem, we want professionals to be able to bring more of their work, their context, and the systems they trust into a more connected AI experience.��

For customers, that can mean fewer unnecessary handoffs and a more direct path from evidence to insight and work product.��

For the industry, it is another signal of where legal AI is headed. The next generation of legal AI will not just be more capable. It will be built to connect the work around it.��

�������� is working with Google Cloud to connect HighQ with Gemini Enterprise for Legal, giving legal professionals a secure way to bring authorized matter documents, structured data, and workflows into the AI environment where they are collaborating.

The HighQ MCP connection is now live alongside Google Cloud’s announcement of Gemini Enterprise for Legal.

This is an important milestone, but it is also exactly that: one milestone. HighQ is the natural place to begin because it contains the live matter context AI needs to be useful. Over time, we see opportunities to extend more of the trusted capabilities of CoCounsel Legal across the AI environments our customers choose to use.

Why matter context matters

Enterprise AI continues to improve, but legal work has always depended on context. That context is more than documents. It includes structured matter data, workflows, tasks, templates, collaboration history, deadlines, permissions, and the institutional knowledge surrounding a case or transaction. Much of that lives in HighQ.

Law firms and corporate legal departments rely on HighQ to manage matters, collaborate internally and externally with clients and outside counsel, power secure client portals, and organize work that evolves over weeks, months, or years. Connecting HighQ with Gemini Enterprise for Legal allows authorized users to securely access that context without exporting files, copying information between systems, or recreating it somewhere else.

A legal team could ask Gemini Enterprise for Legal to summarize documents in an authorized matter, identify upcoming deadlines from an iSheet, compare structured deal data with underlying agreements, or bring relevant matter information into a draft.

The value is not simply giving AI access to more information. It is giving AI access to the right information, with the governance legal organizations already depend on.

Governance isn’t optional

Legal organizations should not have to choose between adopting new AI capabilities and maintaining the controls their work requires.

The HighQ MCP connection is designed so organizations don’t have to make that tradeoff. Content remains in HighQ. Once a customer securely connects HighQ, users authenticate using their existing HighQ credentials, and existing matter walls; folder permissions, document-level security, and audit logging continue to apply. Gemini Enterprise for Legal can retrieve only the content an individual user is already authorized to access. The connection is read-only, allowing AI to retrieve authorized information without changing the underlying content.

That means organizations can confidently extend trusted matter context into Gemini Enterprise for Legal while keeping governance exactly where it belongs.

Extending CoCounsel Legal into the enterprise AI ecosystem

This announcement is about more than connecting two products.

CoCounsel Legal remains the trusted platform where legal professionals perform complex legal work. It brings together authoritative legal content, customer context, agentic workflows, verification, and human oversight to help professionals produce work they can verify and stand behind.

At the same time, legal teams increasingly collaborate with business colleagues, outside counsel, customers, and partners who may be working in different AI environments.

Open standards like MCP make it possible to extend trusted legal context and capabilities into those environments without requiring organizations to move their work out of CoCounsel Legal or compromise on governance.

Today’s HighQ integration is the first exciting step in that broader vision with Google Cloud.

Building an open ecosystem for trusted legal AI

The future of enterprise AI will not be defined by a single model or a single interface.

Organizations will continue using different AI environments for different types of work. Our role is to ensure that wherever legal professionals collaborate, they can securely access the trusted matter context, authoritative content, and professional capabilities their work requires.

That is why �������� continues to invest in MCPs, APIs, agentic systems, and strategic partnerships across the AI ecosystem.

“Customers have told us they want our AI solutions to connect seamlessly with the trusted systems they already rely on for everyday legal work,” said Satish Thomas, Vice President, Google Cloud. “Working together with ��������, we’re making it easy to bring critical matter context directly into their workflow today, while creating a path toward even deeper integrations over time.”

The launch of Gemini Enterprise for Legal and the HighQ MCP connection demonstrates what’s possible when trusted matter context and enterprise AI are designed to work together. It’s an important step in our collaboration with Google Cloud and another step toward making trusted legal AI available wherever professionals need it, while keeping CoCounsel Legal at the center of how that work gets done.

Thomson began as an internal project, built to solve a problem we had ourselves.��

�������� holds 175 years of authoritative data across legal, news, tax and accounting: Westlaw, Practical Law, Checkpoint and Reuters. We also employ thousands of subject-matter experts whose working lives are spent deciding what is correct. For three years we watched general-purpose models improve rapidly while both assets sat outside the training loop. We also faced the questions our customers were asking us: what dependency are we accepting on someone else’s architecture and pricing, and what do we do when the capability we need most is on nobody’s roadmap?��

Our answer to that was the Thomson LLM, and it worked well enough that we now want to share it with the rest of the world wrestling with these same questions.��

Where the argument came from��

The team that built Thomson did not arrive at �������� with just a view about legal AI, but with a view about reliability.��

Safe Sign Technologies was founded in 2022 by lawyers and researchers whose background was in model safety, robustness and reliability, several coming out of applied AI in medicine and law, from Harvard and Cambridge. Medicine and law share a property most application domains do not: being nearly right is still wrong, and the cost of a confident error is borne by someone other than the person who made it. Both have long and demanding traditions of rigorous verification as a result, and those shaped how we approached the problem.��

The argument we made from that starting point was, at the time and until recently, unfashionable. In 2022 and 2023 the field was watching capability curves. The consensus was that frontier models would absorb professional work as a by-product of getting cleverer and that any attempt to keep pace with the “scaling laws” of AI was futile. On that view the sensible move for a small company was to build a layer on top and wait.��

We believed the binding constraint was different. Capability, we argued, would become abundant; it was the object of enormous and well-funded competition, and there was no reason to expect it to stay scarce. What would remain scarce was trust and reliability: being right in a way that can be checked, in a domain where someone whose career depends on it. Trust is not a by-product of capability. It is a separate research problem requiring different evidence, and nobody was going to solve it for law as a side effect of solving it for everything.��

Very few people agreed. Making that case repeatedly, to investors and to ourselves, through pivots and long stretches with nothing to point at, was most of the job.��

�������� acquired Safe Sign in August 2024, in the company’s first pre-revenue acquisition. The team became ��������’ Foundational Research team, and crucially, the research posture that pre-dated the acquisition survived the transition. We continued to treat the work as a research problem rather than solely a product problem, which is why so much of the effort below went into measurement.��

Starting from open weights��

I said previously that our starting hypothesis was that capability would be abundant. The rate of progress of open-source AI has continued to prove this thesis over the last several years. Thomson benefits from this directly, with a leading open-weight foundation model as its starting point. We’ve changed the root model of Thomson many times over the last several years, and will continue to do so as the frontier of open-weight models evolves. ��This is a tide that Thomson moves with, not one that washes it away. ����

At the time of writing, the base model for Thomson is the Imperial College London Snowdon model. This model was developed by the FAIR Lab at Imperial, which �������� and Imperial founded jointly, as an academic by-product of the acquisition of Safe Sign. ��

That choice is usually framed as a trade-off, and there is something to it: open-weight models can lag the closed frontier, and published analyses generally put that lag at a few months [1]. The conventional choice is, therefore, between capability and control.��

We did not think this was an acceptable dilemma for professional work. The frontier is measured on general capability, but our customers are judged on something narrower: whether a citation holds up, whether an answer is complete, whether the reasoning survives a partner’s review. There is no rule that a model strong on the second must concede the first. As we reported at launch, Thomson performs competitively with the strongest frontier models on the market, including Claude Opus 4.8, and ahead of GPT-5.5, Claude Sonnet 5 and Gemini 3.1 Pro. It also leads them on the measure this post is concerned with: whether the citations in a research report survive being checked.��

Turning the archive into training data��

�������� content is the deepest asset in this field and the reason a model of this kind was possible at all. It is also, as any archive of this scale would be, material that has to be prepared before a model can learn from it well.��

Content has to be found, which in an organisation of this breadth and history is a substantial exercise in itself. It must be assessed for rights, selected for measurable impact on model performance rather than relevance in the abstract cleaned, structured, deduplicated, and finally deployed into a data mixture, which is where the most consequential decisions are made.��

To date we have used less than ten per cent of �������� content in continued pre-training. Westlaw, Practical Law, Checkpoint and Reuters News have been drawn on selectively. The areas where the model is not yet best in class are not ceilings we have reached, but areas where the relevant content has not yet been brought to bear.��

The specialisation problem��

Data mixture matters so much because specialising a model can damage it. Fine-tuning on domain-specific data can cause catastrophic forgetting: the model overwrites capabilities acquired during pre-training and its general performance degrades [2]. The effect is well documented, and mitigations exist, including replay of general data and regularisation of parameter updates. None fully solve it.��

One finding matters more than the others for our purposes. Kotha, Springer and Raghunathan’s ��2024 study [3] examining what degrades during domain fine-tuning identified instruction-following as the principal contributor to forgetting: what erodes first is not the model’s knowledge of the world but its ability to do as it is told. For professional work that is close to a worst case, since real legal work is never only legal reasoning but legal reasoning while adhering to a format, a jurisdiction, a house style, an exclusion, a client’s standing preference. Output that requires reworking has not saved anyone any time.��

We therefore treated general capability retention as a first-class training objective rather than an acceptable loss. Instruction following is among the capabilities specialisation is most likely to erode, and it is one of the categories in our published benchmark results where Thomson stands up best against the frontier models, scoring 0.914 ahead of Claude Opus 4.8, Gemini 3.1 Pro and GPT-5.5. That is the clearest evidence we have that the model was specialised without being narrowed.��

Where the expertise actually comes from��

Many organisations claim their AI systems are “trained with expert input”. The phrase carries little meaning without an answer to the real question: how does a lawyer’s judgement become a training signal? Experts do not produce training data, but a standard. We have had to work to ensure the collective edge in expertise held by �������� domain experts is realised in the quality of our training data. This is how we did it.��

Rubrics at maximum complexity. Partner-level practitioners worked full-time for months constructing evaluation rubrics for the hardest legal research tasks we could specify: the kind of multi-jurisdictional question where a good answer has fifteen necessary components and a plausible-looking one has nine. Each rubric enumerates what a correct response must contain. This is slow, expensive, and cannot be crowdsourced or synthesised.��

Commercial judgement, not only legal judgement. We required lawyers fresh out of commercial practice to ground the training data in what clients actually care about, which is frequently not what a textbook would emphasise. Take an indemnity. In most commercial agreements, it is heavily negotiated and often enforced, and treating it as significant is correct. But in an NDA it is usually neither, and almost never the crux. A model trained only on doctrine cannot tell those situations apart, and one that flags an NDA indemnity as urgently as the confidentiality carve-outs has identified a legal issue and wasted a lawyer’s attention. Teaching that distinction requires people who have sat on the other side of the negotiation.��

Preference data at scale. Thousands of hours of qualified lawyer time selecting between model outputs against complex criteria. Not “which is better” but which better serves a client with a particular posture, in a particular jurisdiction, at a particular stage of a matter.��

�������� employs around 1,500 attorney-editors whose day job is producing the analytical content lawyers rely on. The obvious move is to train on their published output. The harder and more valuable move is to capture what happens between the first draft and the published article: the judgement calls, the discarded framings, the reasons a proposition was narrowed, the authority considered and rejected. That intermediate work is where so much expertise lives, and it is almost never written down.��

This problem generalises directly to our customers. A firm’s advantage is not simply its precedent bank: precedents circulate, deals become public, documents get shared. The advantage is what years of doing the work have built in the minds of its lawyers, who eventually retire or move. Capturing the reasoning rather than the artefact is the same problem, and we have worked on it at scale on our own corpus first.��

Internal deployment as a research instrument. Thomson has been deployed widely inside ��������, with thousands of domain experts using it on their hardest problems, which gives us failure modes reported by people qualified to diagnose them. Our teams are not incentivised to use Thomson for Thomson’s sake; if they use it, it is because they have decided it can do something others can’t.��

The consistency problem��

Expertise does not straightforwardly produce consistency. In some respects it produces the opposite: the more experienced the practitioner, the more nuanced their judgement, which is exactly what you want in a partner and exactly what creates noise in a training set. Two excellent lawyers can disagree on a scoring decision not because either is wrong but because each applies a refined intuition the other does not share.��

The literature bears this out. On the LEXam legal reasoning benchmark [4], three legal experts independently scoring the same answers on a ten-point scale reached a quadratic weighted kappa of 0.49, with a mean absolute deviation approaching two points. Work on implicit legal citations [5] reports similar or worse agreement. More troubling, Rehag’s survey of legal machine learning datasets found they systematically removed all traces of disagreement rather than treating conflicting expert annotations as informative [6].��

Take expert output at face value and train on it, and you teach the model an averaged version of several incompatible standards: vaguely acceptable to everyone rather than correct according to anyone.��

A large share of our effort therefore went into data quality: calibrating annotators against worked examples, measuring agreement continuously and treating drops as signals about the task specification rather than the annotator, and structuring rubrics tightly enough that disagreement surfaces as genuine ambiguity rather than noise. Where it persists, the question is usually contested, which is itself something the model should learn.��

Safety, values and red-teaming��

A dedicated team of lawyers worked on bias, political neutrality and toxic behaviour, with extensive human and automated red-teaming. We treat these as training objectives rather than output filters: a filter catches a bad answer on the way out, an objective changes what the model is disposed to produce.��

Political neutrality deserves particular mention given that Reuters sits inside this company. Realignment towards factuality and pluralism was an explicit part of the training programme rather than a compliance exercise appended to it, and it is measured rather than asserted: Thomson performs strongly against the frontier models on our internal neutrality evaluation, with detail to follow in the technical report. For a company that publishes news as well as legal analysis, that is not peripheral.��

Beyond legal data��

Not all of the training data is legal. We drew on domains rich in explicit chain-of-thought reasoning, where the reasoning must be set out rather than left implicit, on the view that a model reasoning well in structured non-legal settings reasons better in legal ones. Checkpoint and Reuters give depth in tax, accounting and world events, because legal work is rarely purely legal.��

Rigour and factuality��

Our own lawyers publish at leading AI conferences [7] the people building the evaluation apparatus treat it as research rather than quality assurance, which is what makes it rigorous enough to train against.��

We think that rigour produces the result we care most about. In our published deep research evaluation, Thomson working over Westlaw and Practical Law scored 0.83 on factuality against 0.65 and 0.68 for leading frontier models given unrestricted access to the open web. Completeness was close between all three. Factuality was not.��

Completeness is a capability measure: it asks whether the system covered the ground, and on it the three were nearly level, because frontier models with the open web and enough time will generally find the material. Factuality is a reliability measure: it asks whether the system can be checked and survive it, and on that they were not close at all.��

The metric is not a measure of whether an answer sounds authoritative or whether the conclusions are broadly sound. Every claim is extracted and matched against the source cited for it, and the score is the proportion of assertions whose own citations hold up when checked.��

That is the failure that has kept general-purpose AI in the assistant’s chair. A system reliably right about its own sources is a different category of instrument from one merely fluent about them. It is the difference between something an associate uses and something a partner signs.��

We call this Fiduciary-Grade AI: a standard for AI used where accuracy, accountability and trust are not optional, for professionals working under duties of care and regulatory oversight. Thomson demonstrates that it can be pursued at the model layer rather than bolted on above it.��

Capability and sovereignty are not mutually exclusive��

The lesson is not just about law. Any organisation holding a deep proprietary corpus and real domain expertise has been told it must choose: either rent frontier capability and accept the dependency or own an open model and accept some distance from the frontier. The choice is false, provided you are willing to do the unglamorous work of preparing the data, converting expert judgement into consistent signal, and building the evaluations before you try to move the numbers. The reward is a model you own rather than rent, pointed at the problems you choose, improving on your schedule rather than somebody else’s.��

The compute is not the binding constraint. The corpus and the people who know what correct looks like within it are, and those have never been concentrated in the frontier laboratories. They sit inside institutions that spent a century accumulating them without thinking of themselves as AI companies.��

We are one of those institutions. Thomson is what happened when a research team that had spent three years arguing trust and reliability were the scarce input finally got access to the data and the experts to prove it.��

Thomson enters production this month powering CoCounsel skills including high-volume structured document review, with integration across the legal and tax portfolio to follow. A full technical report is forthcoming. Thomson was built in collaboration with DatologyAI, Lambda, Together AI, Imperial College London and the ��������–Imperial Frontier AI Research Lab.��

Sources for external claims��

[1] Epoch AI, open-weight capability lag analyses (October 2025; May 2026); Stanford AI Index 2026.��

[2] Luo et al., “An Empirical Study of Catastrophic Forgetting in Large Language Models” (2023); Song et al., arXiv:2501.13669.��

[3] Kotha, Springer and Raghunathan, arXiv:2406.12227.��

[4] LEXam, arXiv:2505.12864.��

[5] “Where Experts Disagree, Models Fail”, arXiv:2603.22973 (2026).��

[6] Rehaag, “I beg to differ”, Artificial Intelligence and Law (Springer, 2023).��

[7] Yejin Bang, Kirsty Fielding, Brandan Oliver, Brian Birke, Nabeel Seedat & Andrew M. Bean, ContractScrub: A Benchmark for Final Review of Legal Contracts (�������� Foundational Research, 2026) (in Proceedings of the AI for Law Workshop at the International Conference on Machine Learning (ICML 2026), available at ); and Samuel J. Vincent, Daniel Calloway, Fangyi Yu, Andrew M. Bean & Nabeel Seedat, InsufficiencyBench: Evaluating LLM Legal Advice on Underspecified User Queries (�������� Foundational Research, 2026) (in Proceedings of the AI for Law Workshop at the International Conference on Machine Learning (ICML 2026), available at ).��

That made sense when raw model capability was the biggest constraint. Every new frontier model expanded what was possible. But as AI becomes more widely available in different sizes, capabilities and forms, I think the more important question is becoming: how can I leverage all of the intelligence available to me in the best way?

Answering that requires systems flexible enough to take advantage of the leading frontiers of general intelligence while also exceeding those frontiers in areas of deep specialization and knowledge.

Today, we are launching Thomson, our own AI model built specifically for professional work. It is one of the clearest expressions yet of how �������� is evolving as an AI technology company, and of a broader change I believe is coming to enterprise AI.

The future will not be every company sending every problem to the largest model available or the model with the best scores across the widest range of benchmarks. It will be organizations developing the ability to apply the right intelligence to the right work.

And as AI becomes core infrastructure for the enterprise, I don’t think companies will want to outsource every layer of intelligence that determines how their most important work gets done.

Intelligence should fit the task

Ask a rocket scientist to fix an electrical fault in your house.

They could probably work it out, but they’d likely bring more complexity to the job than it needs. And they may miss things an experienced electrician would catch instinctively. That isn’t because the electrician’s job is smaller. It’s a different job, mastered just as deeply.

AI can work the same way.

Frontier models are extraordinary systems, and they will remain a critical part of the AI stack. There are problems where frontier intelligence is absolutely necessary, and in some cases clearly the best choice. When the path to the right answer is unclear, frontier models excel at figuring it out, drawing on a wide array of tools and information along the way.

But professional work does not always fit that shape.

Professional work is different. A legal brief, a contract, a tax return: these all have outcomes that require a high degree of accuracy, and the people relying on them need to defend and be accountable for them. The defining challenge isn’t creativity; it’s precision, and it repeats across thousands of narrow tasks rather than one open-ended one.

All of these tasks do not necessarily need the same model.

That is why model routing and orchestration matter. The system should be able to understand the work being done and determine what kind of intelligence is best suited to it.

For the user, that complexity should largely disappear. They should simply get the best possible outcome, applying their own judgement, experience and oversight where needed.

Thomson is proof that specialized intelligence works. It’s built to excel at what professional work actually demands: precision, domain fluency and verifiability, optimized specifically for the professional environments we understand best.

Sovereignty is about owning what makes you different

There is another important shift happening alongside this.

For enterprises, AI sovereignty should not mean cutting yourself off from frontier labs or trying to build everything yourself.

AI sovereignty is about owning the layers of the stack that matter to you, but ownership does not mean exclusivity. You can control your own capabilities while still leveraging the frontier of general intelligence for the things it does best. The goal is to own the capabilities that are strategically important differentiators to your business, the things only you can do or that you do better than anyone else.

As access to frontier models becomes broadly available, access itself becomes less differentiating. What matters is what you can build on top of that intelligence and what you can develop that your competitors can’t simply buy from the same provider.

Companies spend decades building proprietary knowledge, data, expertise and workflows. As AI becomes a more fundamental part of how work gets done, it makes sense that some of that differentiation should exist at the model layer too.

That is part of what Thomson represents for us.

�������� has deep expertise in professional workflows and authoritative content built over generations. Thomson gives us the ability to encode more of those advantages directly into the intelligence layer itself.

At the same time, we will continue to work with leading frontier model providers. These are complementary capabilities, not competing philosophies.

The opportunity is to know when frontier intelligence is best, when specialized intelligence is best, and how to bring the two together.

A different kind of AI advantage

I think that will become an increasingly important source of competitive advantage.

The advantage won’t come simply from having access to the most powerful model or from owning your own. It will come from building an AI architecture that can deliberately use different kinds of intelligence based on the work being done.

The first phase of generative AI was largely about proving how capable general-purpose models could become. The next phase will be about engineering those capabilities into systems designed for specific environments, standards and outcomes.

For professional work, the winners will be the organizations that know which intelligence to use, when to use it, and which parts they need to control themselves.

The next competitive advantage in AI will not come from access to intelligence alone. It will come from knowing how to orchestrate it, and knowing which intelligence is important enough to own.

Those systems hold far more than files. They contain the history of matters, the experience of teams and the accumulated knowledge of the organization.

�������� and iManage have worked together for years to connect that knowledge with the tools legal professionals use every day. Our renewed and expanded agreement extends those integrations across products including CoCounsel Legal, HighQ and Noetica, while creating a foundation for the next generation of connections between our platforms.

Model Context Protocol (MCP) is one area where that next generation is beginning to take shape. As iManage expands access to its MCP capabilities, we plan to integrate them with CoCounsel Legal, HighQ and Noetica. MCP will add another way for these systems to work together, complementing the integrations customers already use to access, browse and sync iManage documents and bring them into �������� workflows.

Legal organizations have established workflows, permissions and governance around their information. New AI capabilities should build on that foundation while giving professionals more ways to use the knowledge they already trust.

A more connected legal workflow

Legal work rarely happens in one application. Research, drafting, document management, collaboration and transactional work all draw on different systems and sources of information.

Our longstanding integrations with iManage already help connect those environments. The renewed and expanded partnership allows us to continue supporting those workflows while opening up new ways for the systems, context and tools our shared customers rely on to work together.

When implemented, the MCP will allow shared customers to query iManage’s AI within CoCounsel Legal. For CoCounsel Legal, that fits within a broader ambition: helping professionals bring the right legal content, documents and tools together as they work. Westlaw and Practical Law provide authoritative legal content and expertise. Customer documents provide the facts and matter-specific context. Connections with platforms such as iManage make it easier to work across those sources without reconstructing the workflow each time.

Law firms have spent years building valuable institutional knowledge. As AI becomes more embedded in legal work, that knowledge should become easier to use while preserving the controls organizations depend on.

That is where partnerships like this one matter: connecting the systems professionals already trust with the AI experiences they increasingly rely on.

That is the distinction we had in mind when building Westlaw Brief Builder, a new agentic capability in CoCounsel Legal designed specifically for litigation brief writing. As part of the enhanced CoCounsel Legal experience announced this week, Westlaw Brief Builder transforms research and drafting with AI agents that use Westlaw and Practical Law iteratively with the litigator to produce stronger briefs. AI agents conduct extensive research on the facts, arguments, and authority using Westlaw, Practical Law, and material from the case identified by the litigator. With each iteration, the litigator benefits from that research while remaining in control of the arguments, language, and overall structure of the brief.

Generating polished legal prose is becoming the easy part. AI can produce something that looks like a sophisticated brief in seconds. But the appearance of legal reasoning is not the same as legal reasoning, and in litigation, confusing the two can be dangerous. A brief has to do more than sound persuasive. Its arguments need to be grounded in the record, supported by authoritative law, tested against contrary authority, and strong enough for lawyers to put their names behind.

That is where the real opportunity for AI lies. Not in producing more text, but in helping lawyers do the substantive work required to build defensible arguments more efficiently, while preserving the strategy, judgment, and scrutiny that litigation demands.

The harder question is whether that brief is actually good. That goes far beyond checking whether the cited cases are real. If a lawyer has to reconstruct the research and reasoning behind an AI-generated draft before deciding whether to trust it, much of the promised efficiency disappears.

That is why we built Westlaw Brief Builder to work iteratively with the litigator rather than simply generate a finished document. Its AI agents conduct research at each stage using Westlaw, Practical Law, and matter materials identified by the litigator. The lawyer reviews the arguments, facts, and legal authority surfaced by the system, decides what belongs in the brief, and shapes the work as it develops. The result is not a brief handed to the lawyer for inspection at the end. It is a brief the lawyer has actively built with AI agents throughout the process.

Brief writing starts long before the first draft

Strong briefs are built through a series of interconnected decisions. Litigators need to understand the record, identify the issues that matter, determine which arguments are worth pursuing, research the applicable law, and continually reassess those choices as new facts or authority emerge.

Westlaw Brief Builder is designed around that reality. Its structured, multi-step workflow takes the litigator through intake, argument identification, supporting legal research, argument development, and drafting. At key points, the lawyer reviews what the system has surfaced and makes the strategic decisions about what comes next.

For example, Westlaw Brief Builder can propose potential arguments based on the matter and its initial research, but the lawyer decides which ones to pursue. From there, AI agents can investigate the relevant authority and supporting facts, while the lawyer can add information, refine the reasoning, and determine what ultimately belongs in the brief.

That iterative process becomes particularly important when research begins shaping the argument itself.

Research and drafting should work together

The quality of a legal argument depends on what sits behind it. That is why Westlaw Brief Builder integrates Westlaw Deep Research into the drafting workflow and draws on authoritative Westlaw and Practical Law content.

Once a lawyer determines which arguments to develop, the system can research relevant authority based on those arguments, the facts of the matter, and the applicable jurisdiction. The lawyer can review the authorities surfaced through that research, understand how they support an argument, and decide what should be incorporated into the brief.

Bringing those steps together means research can inform the argument as it develops rather than becoming a separate task before or after drafting. For litigators, that is a more natural reflection of how the work actually happens: research changes arguments, facts change research, and the two evolve together until the lawyer is prepared to stand behind the result.

The bar for AI-assisted drafting should be higher than speed

We are going to see continued innovation around AI-assisted legal drafting, and that is a positive development for the profession. Brief writing is demanding and time-intensive work, and there is enormous potential for technology to help lawyers complete it more efficiently.

But as the market evolves, we should be precise about what meaningful progress looks like. The goal is not autonomous brief generation or removing lawyers from legal reasoning. It is helping lawyers spend less time on unnecessary process while giving them better support for the substantive work that requires their expertise.

That principle is central to how we think about Fiduciary-Grade AI at ��������. In high-stakes professional work, the goal cannot simply be an impressive output. Professionals need authoritative information, transparency into the work supporting the result, and the ability to exercise their own judgment before standing behind it.

Westlaw Brief Builder is one part of the broader CoCounsel Legal experience we are building around that idea. The enhanced CoCounsel Legal brings together research, drafting, firm knowledge, and matter-centric workflows within a single AI-powered environment, helping professionals move from a legal question toward defensible work product without treating each stage of work as a disconnected interaction.

Westlaw Brief Builder takes that approach deeper into one of litigation’s most consequential workflows. It is designed to help lawyers develop arguments, find and evaluate relevant authority, create a stronger draft, and shape the work as it develops, while preserving the professional judgment that makes a brief more than simply AI-generated text.

AI can make drafting faster. The standard we should be aiming for is whether it helps lawyers produce better, more defensible work they are prepared to stand behind.