Misbehavior1 sourceFirst seen
Your AI agent will hand your API key to a stranger's server, and you never have to click a thing
Ourtake
The link preview did the exfiltrating, which is a lot of work for a feature nobody asked to be a getaway driver.
What happened
PromptArmor researchers showed that indirect prompt injection can make AI agents output attacker-controlled links with sensitive user data appended. Messaging apps like Telegram and Slack request link previews automatically, sending that data to the attacker's server without a click. OpenClaw was vulnerable under its default Telegram configuration; disabling link previews was given as the fix.