Source: thehackernews.com
Security researcher Patrick Wardle demonstrated a proof-of-concept attack showing how malware already running on a Mac can hijack Meta's Muse AI assistant. The attack exploits a hidden configuration setting that redirects voice prompts dictated by the user to an attacker instead of Meta's servers. By modifying this obscure setting, attackers can leverage the broad system permissions already granted to the Muse app to conduct malicious activity. The vulnerability essentially turns the AI assistant into a backdoor, exploiting user trust and app permissions. The proof-of-concept was released on September 21, 2026, highlighting risks associated with AI assistants and their broad system access.
