Because Bots share one identity and machine, untrusted content read by one Bot can reach systems another Bot is authorised for, creating a prompt-injection path into CRM or payment systems.

NotesThis is a risk advisory, not an observed exploit. No source in this sweep documents a successful prompt injection against Grok Bot itself. Mitigation given: treat every external page and message as untrusted input, and separate credentials wherever the product allows it.

If one bot reads untrusted email while another can access a CRM or payment system through the same identity, malicious content may cross the boundary.