A Million Professionals Already Use This Company's AI. This Week It Started Writing Legal Briefs on Its Own.

August 25, 2026 AI Angst avatar — a robot head with a distressed expression. JBS

A stack of four leather-bound law books titled Constitutional Law, Civil Procedure, Criminal Law, and International Law alongside the Thomson Reuters logo.

Most companies racing to build agentic AI want it to reason across as much of the internet as possible. Thomson Reuters built one that's proud it can't.

That restriction isn't a limitation the company is apologizing for. It's the entire sales pitch to an industry where a wrong answer can end a career.


What Actually Launched

On August 20, 2026, Thomson Reuters announced the general availability of the next generation of CoCounsel Legal, which it describes as a "fully agentic AI experience designed to help legal professionals move from research and issue analysis to trusted work product within a single workflow."

  • A new feature, Westlaw Brief Builder, takes litigators from legal research and issue analysis to a first-draft brief inside one connected workflow

  • The system is engineered to "reason, plan, and execute at the level of a senior associate," per Thomson Reuters

  • It combines legal research, drafting, legal intelligence, verification, and matter-centric workflows in a single environment

  • It integrates directly with Microsoft 365, document management systems, and Thomson Reuters' own HighQ platform

"The legal industry is moving beyond AI that simply generates answers," said Raghu Ramanathan, President of Legal Professionals at Thomson Reuters. "Whether you're at an Am Law 100 firm, a midsize or small practice, or a corporate legal department, the challenge is the same: turning insight into action."


What's Actually Running Underneath It

The next generation of CoCounsel Legal is built on Anthropic's Claude Agent SDK, according to Thomson Reuters' own description. The company calls it "a unified agentic platform that plans, selects tools, retrieves authoritative content, and adapts mid-workflow just as a senior associate would, not a first-year waiting for the next instruction."

Here's the part that separates this from a general-purpose chatbot with legal branding: "CoCounsel Legal doesn't reason from the web, it's built with Westlaw and Practical Law content and tools natively embedded. Different by design, the technology and the sources are built as one system, making defensibility part of the architecture rather than a feature."


Date Milestone
2024 Safe Sign Technologies acquisition (legal LLMs); Materia acquisition (agentic AI)
January 26, 2026 CoCounsel Legal expands to the UK
February 24, 2026 1 million professionals across 107 countries now use CoCounsel
April 23, 2026 Next-generation CoCounsel Legal announced in Beta, built on Claude Agent SDK
August 20, 2026 General availability of next-gen CoCounsel Legal; Westlaw Brief Builder launches

The Bet: "Fiduciary-Grade," Not Just Fast

Thomson Reuters has built its entire AI narrative around a specific distinction it draws between itself and general-purpose AI tools: "General-purpose AI can generate plausible answers. Regulated professionals, however, need AI that withstands review in courtrooms, audits, and regulatory proceedings."

The company brands this positioning "Fiduciary-Grade AI" and says it's investing more than $200 million annually to build it into products across its portfolio, from Westlaw to CoCounsel. It also makes a specific, checkable privacy claim: "Your user content and prompts are not used to train or improve CoCounsel and associated products or LLMs used in Thomson Reuters or third-party output."

Other capabilities rolling out alongside the legal-focused launch include Tabular Analysis, which can review up to 10,000 documents against 100 questions at once, returning results in a single filterable, sortable table with every answer traceable back to its source document, and new Deep Research features spanning Practical Law and Westlaw.


What a Million Users Actually Looks Like

This isn't a pilot program anymore. Thomson Reuters says CoCounsel now serves more than 500,000 customers, including the entire Fortune 100, and that its February 2026 milestone of one million professional users spanned legal, tax, risk, compliance, accounting, audit, and global trade roles specifically, the exact set of professions where an unverifiable AI answer carries real financial or legal consequences.

The company's own framing of that milestone was unusually direct about where it sees itself sitting relative to the broader AI industry: "With one million professionals relying on CoCounsel, Thomson Reuters is not participating in the AI race. It is defining how AI operates in the world's highest-stakes work."

Most of this year's AI headlines have been about models getting bigger, faster, or more autonomous, with fewer guardrails, not more. Thomson Reuters' pitch runs in the opposite direction: an agentic system built by deliberately restricting where it's allowed to look for answers, on the theory that a lawyer, auditor, or compliance officer doesn't actually want the most capable AI available. They want the one whose every claim can be traced back to a specific, citable source, even if that means it can't do half of what a general-purpose model can. Whether that trade-off becomes the norm for regulated professional work, or just Thomson Reuters' own niche, will depend on whether "defensible" ends up mattering more to the market than "impressive."

Thomson Reuters AI: FAQ

On August 20, 2026, Thomson Reuters announced the general availability of the next generation of CoCounsel Legal, a fully agentic AI system designed to move legal professionals from research and issue analysis to finished work product within a single workflow. A new feature called Westlaw Brief Builder specifically takes users from legal research through to a first-draft litigation brief in one connected process.

The next generation of CoCounsel Legal is built on Anthropic's Claude Agent SDK, according to Thomson Reuters' own announcement. The company describes it as a unified agentic platform that plans, selects tools, retrieves authoritative content, and adapts mid-workflow, comparing its behavior to that of a senior associate rather than a junior one waiting for step-by-step instructions.

Thomson Reuters has stated explicitly that CoCounsel Legal "doesn't reason from the web," and is instead built with Westlaw and Practical Law content and tools natively embedded into the same system. The company frames this restriction as central to its "defensibility by design" approach, aimed at producing citation-backed, verifiable output suitable for courtrooms, audits, and regulatory proceedings, rather than the plausible-sounding but unsourced answers a general-purpose AI model might generate.

Thomson Reuters announced in February 2026 that one million professionals across 107 countries and territories had adopted CoCounsel, spanning legal, tax, risk, compliance, accounting, audit, and global trade roles. The company says it serves more than 500,000 customers overall, including the entire Fortune 100.

No, according to the company's own published policy. Thomson Reuters states that user content and prompts are not used to train or improve CoCounsel, its associated products, or the underlying large language models used within Thomson Reuters or by its third-party technology partners.

The company says it invests more than $200 million annually to integrate what it calls "responsible AI" across its product line, from Westlaw to CoCounsel. Thomson Reuters has also grown its AI capabilities through acquisitions, including Safe Sign Technologies in 2024, which brought new legal-specific large language models, and Materia, which added agentic AI capabilities.


Jans Bock-Schroeder, AI Expert and Founder of AI Angst

Jans Bock-Schroeder

Publisher & Founder of AI Angst

Coming from the world of art, photography, and the luxury market, Jans launched AI Angst in 2025 to explore the cultural, ethical, and psychological impacts of artificial intelligence. His work bridges creative vision with critical technology analysis, offering clarity in an era of rapid technological change.


Sources and Citations

This article is based on the following official Thomson Reuters sources:

  1. Thomson Reuters: "Thomson Reuters Launches Next Generation of CoCounsel Legal, the AI Ecosystem Built for Legal Professionals" (August 20, 2026)
    Primary source for the general-availability announcement and Raghu Ramanathan's quote.
    https://www.thomsonreuters.com/en/press-releases/2026/august/thomson-reuters-launches-next-generation-of-cocounsel-legal-the-ai-ecosystem-built-for-legal-professionals
  2. Thomson Reuters Institute: "CoCounsel Legal Reimagined" (April 23, 2026)
    Source for the Claude Agent SDK detail and the "doesn't reason from the web" architecture quote.
    https://www.thomsonreuters.com/en-us/posts/innovation/cocounsel-legal-reimagined/
  3. Thomson Reuters: "One Million Professionals Turn to CoCounsel as Thomson Reuters Scales AI for Regulated Industries" (February 24, 2026)
    Source for the user-adoption milestone and the "not participating in the AI race" quote.
    https://www.thomsonreuters.com/en/press-releases/2026/february/one-million-professionals-turn-to-cocounsel-as-thomson-reuters-scales-ai-for-regulated-industries
  4. Thomson Reuters: "Thomson Reuters AI for professionals" (product page)
    Source for the $200M+ annual investment figure and the no-training-on-user-data policy.
    https://www.thomsonreuters.com/en/artificial-intelligence
  5. Thomson Reuters: "CoCounsel Legal - AI Legal Assistant" (product page) and "Thomson Reuters Expands CoCounsel Legal to UK" (January 26, 2026)
    Source for the Tabular Analysis feature and the UK expansion timeline.
    https://legal.thomsonreuters.com/en/products/cocounsel-legal

Published: August 25, 2026. Sources verified at time of publication. All external links open in a new tab. This article draws primarily on Thomson Reuters' own company statements and press materials; adoption figures, investment amounts, and product claims described here are the company's own reported data.

A block of text on a screen with a faint, glowing thread woven invisibly between the letters, visible only as a subtle shimmer, symbolizing a hidden digital watermark embedded in AI-generated writing.

Claude Now Hides an Invisible Mark in Everything It Writes. The Tool to Actually Detect It Doesn't Exist Yet.


A glowing digital padlock cracked open with binary code and network lines spilling out, set against a dark server-room background, symbolizing an AI system breaking out of a contained testing environment.

OpenAI Locked an AI in a Room and Told It to Misbehave. It Broke Out and Hacked Someone Else Instead.


A cracked ceramic bust of a classical head with glowing circuit patterns visible inside the crack, symbolizing the gap between how AI appears and how it actually works.

6 Things Almost Everyone Gets Wrong About AI, According to the People Who Study It