On Tuesday, Anthropic launched a brand new file-creation characteristic for its Claude AI assistant that permits customers to generate Excel spreadsheets, PowerPoint shows, and different paperwork instantly inside conversations on the internet interface and within the Claude desktop app. Whereas the characteristic could also be helpful for Claude customers, the corporate’s help documentation additionally warns that it “might put your information in danger” and particulars how the AI assistant will be manipulated to transmit person information to exterior servers.
The characteristic, awkwardly named “Upgraded file-creation and evaluation,” is mainly Anthropic’s model of ChatGPT’s Code Interpreter and an upgraded model of Anthropic’s “evaluation” instrument. It is at present accessible as a preview for Max, Workforce, and Enterprise plan customers, with Professional customers scheduled to obtain entry “within the coming weeks,” in keeping with the announcement.
The safety challenge comes from the truth that the brand new characteristic offers Claude entry to a sandbox computing setting, which permits it to obtain packages and run code to create recordsdata. “This characteristic offers Claude Web entry to create and analyze recordsdata, which can put your information in danger,” Anthropic writes in its weblog announcement. “Monitor chats intently when utilizing this characteristic.”
