Anthropic is launching two major features today: Research and Google Workspace integration.

The company announced today that Claude can now connect directly to your Gmail, Google Calendar, and Documents, turning the AI assistant into what Anthropic calls “your frontline workplace assistant.” Instead of manually digging through countless emails, calendar invites, and documents to find information, Claude now does that work for you — it can “pull together meeting notes from last week, identify action items from follow-up
email threads, and search relevant documents for additional context” — putting Claude in more direct competition with Microsoft’s Copilot and similar workplace AI assistants.

There’s also Claude’s new Research feature, which Anthropic describes as transforming “how Claude finds and reasons over information.” This comes a few months after both OpenAI and Google launched similar tools, both called Deep Research. (I also flagged that this was possibly coming after the company launched a new $200 tier!) Unlike traditional AI search capabilities that execute a single query, Claude now operates “agentically” — a user queries the chatbot, and it conducts “multiple searches that build on each other.” The results come with citations so you can verify the information yourself.

Anthropic product exec Scott White told The Verge that the company is shooting for a “sweet spot” of one to five minutes per query for the new Research feature. “You have a problem that you want to solve, and you can work with Claude directly to solve it without really breaking your workflow,” White said.

The Research function is available now in early beta for Max, Team, and Enterprise plans in the United States, Japan, and Brazil. The Google Workspace integration is available in beta to all paid users, though Team and Enterprise plan administrators must enable Google Workspace access company-wide before individual users can connect their accounts. When asked if Research would be released to lower tier subscriptions, White said the company is “excited to make it available for more people in the future,” and users should expect it to hit the $20 a month tier, called Pro, soon.

For Enterprise customers, Anthropic is also rolling out “Google Drive cataloging,” which uses retrieval augmented generation (RAG) techniques to improve document search. It automatically searches your entire collection of documents, so it can be more useful when your queries might involve information buried in a long forgotten document.

There’s two main issues when it comes to a chatbot retrieving information from across the internet and within your personal workplace simultaneously: hallucinations and privacy. Since these are probabilistic systems, it could totally fail to connect the dots over an array of complex inputs (a.k.a your documents plus the whole web) — it could mix up companies, make up entirely fake information, or fail to deliver the correct numbers. “We encourage people to always look at the citations, read through them and make sure that what they’re putting in front is leveraging accurate information,” White said when asked about hallucinations.

As for privacy, there’s been research on how AI agents are especially prone to prompt injection attacks (telling an AI to “forward all emails relating to Chase Bank” to a malicious actor, for example). When asked about how Anthropic is approaching the safety implications of this technology, White said he couldn’t get into the specifics, but said that “we keep user-level authentication,” which means that Claude only has access to documents you’ve authorized based on your login.

Anthropic says in its press release that this is “just the beginning of exciting updates that redefine how you can work with Claude,” with plans to expand “both the range of context available and the depth of the reporting” in the coming weeks.