Anthropic today introduces two main functions: Google’s research and integration.
Today, the company announced that Claude can now connect directly with your Gmail, Google calendar and documents, transforming AI assistant into what anthropic calls “assistant workplace on the first line.” Instead of manual digging countless emails, calendar invitations and documents to find information, Claude now it works for you-I can “combine notes from last week, identify elements of activities from further
E -mail threads and search the relevant documents in search of an additional context ” – introducing Claude in a more direct competition with Copilot Microsoft and similar AI assistants in the workplace.
There is also a novel Claude research function, which Antropic describes as a transformation “how Claude finds and reasons for information.” This is a few months after both Openai AND Google I launched similar tools, both called deep research. (I also marked that it probably took place after the novel USD 200 level was launched!) As opposed to the established AI search options, which perform one question, Claude is now “aggressive” – the user asks Chatbota and conducts “many searches that will develop”. The results are delivered with quotes so that you can verify the information yourself.
Scott White said the anthropic Scott White product The Verge that the company turns a “sweet place” from one to five minutes to ask for a novel research function. “You have a problem you want to solve, and you can work with Claude directly to solve it without breaking the flow of work,” said White.
The research function is now available in the early beta version for Max, Team and Enterprise plans in the United States, Japan and Brazil. The integration of Google working space is available in beta for all paid users, although the administrators of teams and enterprises must enable Google Workspace Access to the whole company before individual users can connect their accounts. Asked if the research would be issued in relation to the subscription to lower levels, White said that the company is “excited to make them available more than the future”, and users should expect that it will soon reach USD 20 per month, called PRO.
In the case of corporate clients, Anthropic also introduces “cataloging Drive Google”, which uses the techniques of downloading generation (RAG) to improve document search. Automatically searches the entire set of documents, so it can be more useful when queries may include information buried in a long -forgotten document.
There are two main problems when it comes to chatbot, which downloaded information from the Internet and in his personal workplace at the same time: hallucinations and privacy. Because these are probabilistic systems, it may not completely connect dots through a number of elaborate input data (i.e. your documents and the entire network) – it can mix companies, create completely false information or not provide the correct numbers. “We encourage people to always look at quotes, read them and make sure that what they put in the front, uses accurate information,” said White, when asked about hallucinations.
As for privacy, research was conducted on how AI agents are particularly susceptible to swift injection attacks (telling AI about “passing all E -Mailes about Chase Bank”, for example, a malicious actor). When asked about how Anthropic is approaching the consequences of the safety of this technology, White said that he could not get to the details, but said that “we are conducting authentication at the user level”, which means that Claude has access only to documents that you authorized on the basis of logging in.
Anthropic says in a press release that this is “just the beginning of exciting updates that will redefine the way of working with Claude”, with plans to expand “both available context and depth of reporting” in the coming weeks.