Anthropic is turning Claude into a legal workflow engine

Anthropic's latest release is doing two things at once. One is obvious: it is shipping a more capable model. The other is more revealing: it is shipping a more governable one.

Claude Opus 4.7 is described in Anthropic's docs as its most capable generally available model to date, aimed at long-horizon agentic work, knowledge work, vision tasks, and memory tasks. The page highlights a 1M-token context window, high-resolution image support, a new xhigh effort level, and beta task budgets. It also says adaptive thinking is the only thinking-on mode, that thinking content is omitted by default, and that the model is more literal about instruction following. Those are not just features. They are controls.

That matters because the company is now pushing the model into a place where controls matter more than charm. Reuters reported that Anthropic is expanding Claude's AI tools for law firms and lawyers, including specialized legal topics and access to other legal research and AI products. Thomson Reuters then made the pairing explicit: it announced an MCP integration that connects Claude directly to CoCounsel Legal, and said the next generation of CoCounsel Legal is being rebuilt on Anthropic's Claude Agent SDK. Lawyers will be able to move between general-purpose AI and citation-grounded legal work from either environment.

Thomson Reuters is not pretending the legal market is a casual environment. In its own framing, CoCounsel Legal is supposed to be fiduciary-grade AI, built around accuracy, accountability, and trust. The company says it reasons across 1.9 billion Westlaw and Practical Law documents, 1.4 billion KeyCite validity signals, and a citation ledger that makes every source traceable in one click. It also says customer data is not used to train third-party models and is not shared beyond the customer's own environment. Those are all vendor claims, but they are the right sort of claims for this market.

Legal work is a good test for that sort of thing because it has little patience for vibes. Confidentiality matters. Citation provenance matters. Review matters. The wrong answer is not merely embarrassing; it can be sanctionable, expensive, or both. A system that wants to live there has to do more than sound helpful. It has to show its working.

That is why the most interesting part of Anthropic's current packaging is not raw capability but bounded delegation. A model that can read more context, inspect high-resolution images, and reason against a task budget is a different product class from a model that simply answers quickly and confidently. The former can be slotted into a workflow with checkpoints. The latter is just a very fluent liability.

There is still a gap between the announcements and the field. Reuters' report is still mostly a partnership and product story, not proof of broad law-firm adoption. Thomson Reuters' language is, naturally, vendor language: fiduciary-grade AI, validated references, trust, accountability. Useful words, all of them, but not the same thing as independent operational proof. What is clear is the direction. What is not yet clear is the adoption curve.

The useful conclusion is a plain one. Frontier model vendors are now competing on whether their systems can survive inside professions that care about logs, citations, confidentiality, and review. That means the model is only half the product. The other half is the permission structure around it. In legal work, confidence is cheap. Receipts are the product.

Source basis

• Anthropic: “What’s new in Claude Opus 4.7” — official product documentation on the 1M-token context window, high-resolution image support, xhigh effort, task budgets, adaptive thinking, and related behavior changes.
• Reuters: “Anthropic expands Claude's AI tools for law firms, lawyers” — independent reporting on the legal-tools expansion and access to specialized legal topics plus other legal research and AI products.
• Thomson Reuters: “Thomson Reuters and Anthropic Expand Partnership to Connect Claude with CoCounsel Legal” — partnership announcement describing the MCP integration, fiduciary-grade framing, source traceability, and the next-generation CoCounsel Legal workstream.
• TechCrunch: “The AI legal services industry is heating up — Anthropic is getting in on the action” — secondary coverage noting the expansion of Claude for Legal and the addition of legal plug-ins and MCP connectors.