I'd love to know if anyone has concrete ideas on how to protect against prompt injection. It's clearly a major vulnerability, any ideas out there on approaches

I'd love to know if anyone has concrete ideas on how to protect against prompt injection. It's clearly a major vulnerability, any ideas out there on approaches that could scale to protect against this?