Data Exfil from Agents in Messaging Apps

Communicating with AI agents (like OpenClaw) via messaging apps (like Slack and Telegram) has become much more popular. But it can expose users to a largely unrecognized LLM-specific data exfiltration risk, because these apps support ‘link previews’ as a feature. With previews enabled, user data can be exfiltrated automatically after receiving a malicious link in an LLM-generated message -- whereas without previews, the user would typically have to click the malicious link to exfiltrate data. F…