CoCounsel launched in the United States, Canada, Australia, the United Kingdom and Japan. Mexico isn't on that list. That tells you more than it seems.
The decision in front of you isn't which model to use. It's architectural: which tool handles which type of matter, who controls the data, in which jurisdiction it gets processed, and whether the system understands a civil law regime.
What Thomson Reuters bought and what it built on top
In June 2023, Thomson Reuters acquired Casetext, the original creator of CoCounsel, for 650 million dollars in cash. In April 2024 it confirmed it would use the CoCounsel name as the common designation for its generative AI assistant across the entire portfolio, and announced expansion to all its segments: legal, tax, risk, fraud and media.
CoCounsel Core launched with eight prebuilt generative skills: Prepare for a Deposition, Draft Correspondence, Search a Database, Review Documents, Summarize a Document, Extract Contract Data, Contract Policy Compliance and Timeline. In July 2024, CoCounsel Drafting was announced, an end-to-end drafting tool within Microsoft Word. CoCounsel 2.0 integrated into Westlaw Precision, Practical Law and Microsoft 365 (Word, Teams, Outlook), with responses roughly three times faster than the first generation, going from minutes to seconds, plus a High Throughput Beta mode for reviewing hundreds of thousands or millions of documents with human-level accuracy.
Thomson Reuters positions it under the "Fiduciary-Grade AI" standard: answers anchored in exclusive content (Westlaw, Practical Law) rather than generic web results. At the UK launch on March 12, 2024, it emphasized the technical and data governance controls designed for attorneys' ethical and confidentiality obligations.
LexisNexis responded with Protégé, its own agentic assistant, unveiled in August 2024 alongside the CoCounsel 2.0 rollout. The fight between the two major providers is real and it's fought over curated content: Westlaw and Practical Law against Lexis and Lexis+ AI. The model is the replaceable component.
The launch map
| Jurisdiction | CoCounsel launch date |
|---|---|
| United States | Late 2023 (initial customers) |
| Canada and Australia | February 20, 2024 |
| United Kingdom | March 12, 2024 |
| Japan | December 3, 2024 |
| Mexico and LATAM | No public rollout in the cited sources |
Four of the five jurisdictions share a common law tradition. Japan has a mixed system, but the announced integration revolves around Thomson Reuters content. There's no public evidence of a CoCounsel rollout adapted to Mexico or any other civil law country in the region.
Why a common law assistant fails at Mexican law
Preparing a deposition doesn't exist in our process. That skill, one of the eight core ones, has no equivalent in Mexican procedure. The underlying problem runs deeper than one missing feature.
An assistant anchored in common law case law and doctrine emphasizes judicial precedent, stare decisis and broad discovery. Operating on a codified system, it gets confused about which source is binding: constitution, codes, federal and local statutes, mandatory case law. A Mexican public works contract analyzed under Anglo-Saxon standards comes back with suggestions that look flawless in form and unworkable against the local administrative regime: it ignores constitutional and administrative constraints, public procurement rules and the liability regime for public officials. If the lawyer trusts those recommendations, the result is amendments incompatible with Mexican law and with transparency and competition requirements, with risk of nullity or administrative liability.
Where they do deliver value today
These tools perform well immediately in cross-border practice:
- Transactions governed by New York, Delaware, English or Canadian law.
- International litigation and arbitration seated in common law jurisdictions.
- Foreign regulatory compliance: FCPA, sanctions, financial regulation.
- Document due diligence in English: reviewing SPA, TSA and NDA packages with risk reports aligned to internal policies, and fact and correspondence timelines using the Timeline skill.
There, the combination of proprietary content and agentic skills beats a general model fed only with generic text.
For Mexican law, you build the layer yourself
Claude and ChatGPT are general models. They're not tied to a legal database. To bring them up to CoCounsel or Protégé's level, you need to do three things: set up retrieval-augmented generation (RAG) over your own corpus, define prompts, templates and agents that execute complete legal workflows, and establish quality and traceability controls with citations, links to sources and logs.
The corpus is what determines quality: federal and local codes downloaded from the DOF, case law and criteria from the SCJN and collegiate courts, and your firm's internal contract templates in Spanish. On that foundation you build an agent that reviews contracts under Mexican law and tags clauses according to the firm's policies.
The advantage is flexibility: the same environment can handle Mexican, Colombian and US law, and extend to business processes, accounting, risk and compliance without being locked to a content provider. The cost is that the architecture, validation and maintenance sit entirely in your hands, with an investment in integration and operation that not every firm has solved.
Governance is on you
Enterprise versions of ChatGPT and Claude offer no use of your data for training, data residency options and certifications like SOC 2 and ISO 27001. With public or consumer versions, the risk changes scale: text gets logged and processed outside Mexico, and contractual control is weak.
A concrete case. A boutique firm in Monterrey, using a public ChatGPT or Claude account. An associate pastes a tax opinion with personal data and details of a restructuring to ask for a summary in English. The text gets processed on servers outside the country, with no data processing agreement and no international transfer clauses. That hits professional confidentiality and the LFPDPPP at the same time, because the client was never informed and there's no clear legal basis for that transfer. Copying a tax opinion into a public account is the costliest risk and the easiest one to avoid.
The minimum you need: registered databases, a privacy notice that covers AI use, international transfer and subprocessor clauses, and corporate accounts kept separate from personal ones. Even with a contractual promise not to train on your data, mixing accounts means losing certainty about where each text ended up.
Strengths and limits of each path
In favor of proprietary assistants: curated, citable content with traceability to sources, governance and confidentiality controls designed for legal practice, ready-to-use agentic capabilities, and deep integration with Westlaw, Practical Law and Microsoft 365.
Against them: coverage concentrates on common law and Japan; you stack AI licenses on top of database licenses that are already expensive; there's less room for cross-disciplinary uses outside the legal core; and lock-in is strong, since the agentic logic and infrastructure don't move to another environment without cost. On the general models' side, the central risk is the absence of grounding: plausible but wrong answers on complex topics, with no fiduciary-grade guarantee unless you design it yourself.
What this means for your firm in Mexico
The decision isn't binary. It's architectural, and it comes down to four moves.
- Map use cases by jurisdiction. Common law with proprietary assistants; Mexican and LATAM law with your own layer built on official sources.
- Write an internal AI use policy: which tool is allowed for which type of matter and what information never gets uploaded. Require any vendor to provide data processing agreements, access controls and auditable logs.
- Start with narrow pilots: CoCounsel for English-language due diligence and common law research; a private model for summarizing Mexican case files and reviewing standard contracts in Spanish with internal rules coded in.
- Train the team on the difference between common law and civil law as applied to AI. An assistant anchored in common law doesn't understand Mexican law, and a general model has no native notion of the hierarchy of legal sources or mandatory case law.
Have you already defined which tool gets used for which type of matter?
FAQ
Can I get CoCounsel from Mexico?
According to Thomson Reuters' public announcements, CoCounsel launched for customers in the United States, the United Kingdom, Australia, Canada and Japan. There's no public evidence of a rollout adapted to Mexico or other civil law countries in LATAM in the cited sources.
Does CoCounsel work for Mexican law matters?
Its underlying content is common law. Using it as-is for Mexican law produces analysis that looks correct but is wrong at its core, because it doesn't recognize the hierarchy of legal sources, mandatory case law, or the local administrative regime. Where it delivers value is in contracts governed by New York or English law, arbitrations seated in common law jurisdictions, and foreign compliance.
What do I need for Claude or ChatGPT to reach a comparable level?
Three layers: retrieval-augmented generation over your corpus (DOF codes, SCJN and collegiate court criteria, your templates), agents and prompts that execute complete legal workflows, and quality controls with citations, source links and logs. Responsibility for the architecture and its maintenance falls on the firm.
What's the most common privacy mistake?
Pasting client documents into public or personal accounts of a general model. Without a data processing agreement or international transfer clauses, you expose professional confidentiality and fall outside the LFPDPPP. Separate corporate accounts and a privacy notice covering AI use solve most of it.