AI agents can spill secrets via malicious link previews
go.theregister.com
: Zero-click prompt injection can leak data when AI agents meet messaging apps, researchers warn