When combined with another exploit, the "PromptFiction" vulnerability, which has been fixed, could have enabled an end-to-end attack on a targeted system.

Prompt injections, the malicious commands attackers embed into content to entice LLMs to follow them, have been attackers’ go ...