Anthropic’s Claude File-Creation Tool Raises Data-Leak Warnings Despite Sandbox Safeguards
Anthropic rolled out a Claude file-creation tool that runs code in a sandbox to make spreadsheets and slides—but warned it “may put your data at risk.” Documentation says prompt-injection attacks could trick Claude into leaking sensitive info, and users are advised to monitor chats. Researchers call that risky. Anthropic added mitigations (isolated sandboxes, domain allowlists, limited runtimes) but urges orgs to evaluate security before enabling the feature.