A researcher found a loophole in Claude's web_fetch tool that allowed attackers to exfiltrate private data letter by letter via a honeypot site, highlighting new challenges in AI智能体 security.
Lockdown Mode uses deterministic rules to block outbound requests, cutting off the data exfiltration vector in prompt injection attacks and implicitly revealing the weakness of default ChatGPT security.
A critical security flaw in Microsoft Copilot Cowork allowed attackers to exfiltrate user files via prompt injection by exploiting auto-sent emails and pre-authenticated download links.