Anthropic embeds Claude in Word
Anthropic rolled out a beta of “Claude for Word,” embedding its assistant inside Microsoft Word with features like built-in editing tools and cited answers for professional users (e.g., legal and finance). (eweek.com) Banks have begun testing Anthropic’s new model, Mythos, even as users have reported performance issues—especially around Claude Code—which has prompted a wave of complaints about reliability. (siliconrepublic.com) (aol.com)
Anthropic has put Claude directly inside Microsoft Word, opening a new front in the fight for office software as banks test a separate high-risk model from the same company. (businessinsider.com) The Word product is a beta add-in for Microsoft Word aimed at Team and Enterprise users, with legal contract review listed as an early use case. Reports say it can make tracked edits, answer questions with citations, and work alongside Anthropic’s add-ins for Excel and PowerPoint. (thenextweb.com) (thurrott.com) That puts Anthropic inside the document where many lawyers, bankers, and consultants already draft contracts, memos, and board materials, instead of sending text out to a separate chatbot window. Business Insider described the move as a direct challenge inside Microsoft’s own software stack. (businessinsider.com) The timing overlaps with a more sensitive Anthropic rollout: banks in the United States, Canada, and the United Kingdom have begun testing Claude Mythos Preview while regulators weigh cybersecurity risks. Silicon Republic reported those tests on April 13, one day after Reuters-style coverage surfaced meetings between authorities and large banks. (siliconrepublic.com) (msn.com) Mythos is not a writing helper for Word. Anthropic’s own system card, published April 7, says the model showed unusually strong performance on cyber tasks, including identifying thousands of high- and critical-severity vulnerabilities in testing, which is why access is being tightly limited. (anthropic.com) (usnews.com) That split screen matters for buyers. Anthropic is pitching Claude deeper into everyday office work at the same moment it is asking governments, banks, and security teams to treat one of its newest models as powerful enough to require extra controls. (anthropic.com) (siliconrepublic.com) The company is also facing complaints from developers who say Claude has become less reliable in recent weeks, especially in coding tasks. AOL reported April 14 that users described regressions in Claude Code and in Anthropic’s flagship Claude models, while VentureBeat reported similar claims and said Anthropic leaders pushed back on the idea that the core model had simply been “nerfed.” (aol.com) (venturebeat.com) Separate security issues have added to that scrutiny. TechSpot reported last week that a Claude Code package on npm was published with a large JavaScript source map, and other recent coverage pointed to a disclosed flaw that could let a malicious Git configuration trigger code execution before Claude Code’s trust prompt appeared. (techspot.com) (msn.com) For Microsoft, the Word add-in also creates an awkward overlap. Anthropic is selling an assistant inside the same word processor where Microsoft already sells Copilot, turning Word into a market for competing artificial intelligence tools rather than a single built-in assistant. (msn.com) (businessinsider.com) Anthropic’s bet is that professional users will keep asking for more help exactly where they write. The harder question, as Word customers and bank testers weigh these products, is whether the company can expand Claude’s reach faster than doubts about reliability spread. (thenextweb.com) (aol.com)
