San Francisco, April 13, 2026, 05:10 PDT
Anthropic launched Claude for Word in beta over the weekend, putting its AI assistant inside Microsoft Word as a sidebar that can answer questions about a document, rewrite selected text and route every change through Word’s native tracked-changes system so users can accept or reject each revision. On Anthropic’s own help page, legal contract review is the first example use case it lists.
The release matters because it pushes Anthropic deeper into one of AI’s busiest enterprise battlegrounds: legal work. Law firms are already under pressure to adopt the technology, and McCarthy Tetrault partner Susan Wortzman told Reuters last month that firms once had to reassure clients they were not using generative AI, but “now they are saying you must use it.” Reuters
It also lands in Microsoft’s core software territory. Microsoft said in late March that its Researcher agent in Microsoft 365 Copilot can work with Claude models, and Microsoft executive Nicole Herskowitz told Reuters that customers want “various different models from different vendors” inside Copilot; Anthropic is now offering its own Word-native route alongside that relationship. Microsoft Support
Anthropic says the add-in is available to Team and Enterprise customers and works on Word for the web, Windows and Mac. Team seats start at $25 per user per month billed monthly, Anthropic’s pricing page shows, and users can switch between Sonnet 4.6 and Opus 4.6 while working through comment threads or drafting into existing templates without breaking numbering and styles.
The legal pitch is not subtle. Anthropic’s sample prompts include asking Claude to summarize governing law and other commercial terms, flag provisions that are off-market, or outside standard deal terms, make indemnity language, clauses that allocate liability, mutual, and explain which counterparty redlines, edits marked for review, are dealbreakers.
That puts Claude for Word up against specialist legal AI players such as Harvey and Legora, and against incumbents including LexisNexis and Thomson Reuters that bundle AI with proprietary case-law and workflow products. Reuters reported last week that Harvey was valued at $11 billion after a new funding round, Legora at $5.55 billion after its own raise, and that both are courting law students as they chase long-term demand in a legal AI market some estimates put on track to reach $10 billion a year by 2030.
The risk is that speed does not remove the need to check the work. Hallucinations, when AI systems invent facts or citations, have already led courts around the United States to question or discipline lawyers in at least seven cases, Reuters reported in February, and Suffolk University Law School dean Andrew Perlman said using generative AI to create citations without checking them is “incompetence, just pure and simple.” Reuters
Anthropic has had its own taste of that problem. Reuters reported last year that a Latham & Watkins lawyer defending the company in a copyright case told a federal judge Claude had produced a wrong article title and author names for a real source, prompting the judge to call it a “very serious and grave issue.” Reuters
Anthropic’s own documentation flags another uncertainty: prompt injection, or hidden instructions buried in text, comments or tracked changes that can try to steer the model into doing something unintended. The company warns against using the add-in on untrusted outside documents and says the beta is not recommended for final client deliverables or litigation filings without human review; chat history is not saved between sessions, and the feature is not yet covered by Enterprise audit logs or its Compliance API.
Claude for Word also shares context with Anthropic’s Excel and PowerPoint add-ins, a sign that the company wants Claude to follow office workers across documents, spreadsheets and slide decks rather than sit in a separate chat window. Anthropic said in March that it had upgraded its Excel and PowerPoint add-ins to share conversation context across apps, and its Word help page says the same workflow now extends across all three.
For law firms, the question now is not whether AI shows up inside Word, but how much of the job they want it to do there. Anthropic is betting convenience inside the document can win work away from specialist vendors, but the need for verification, proprietary legal content and tighter controls could still leave room for the specialists.
