AI & Technology

A critical vulnerability dubbed PromptFiction in Claude Desktop allowed hackers to take control of victims' AI with a single click. Disclosed by Oasis Security, the breach exploited the system's "claude://" shortcut to force the artificial intelligence to execute commands automatically, without any user approval. The attack used disguised links containing malicious instructions that remained hidden within Claude's own interface. This enabled attackers to silently steal conversation histories or corporate data and transmit them to their own servers. In more severe cases, the vulnerability also allowed malware injection and full remote control of the infected computer. Flaw in Claude Desktop Allowed Data Theft via Hidden Instructions