Gandalf chatbot security game counters privacy fireballs

You shall not pass judgement, Lakera AI insists, because exposed player info was harmless

Gandalf, an educational game designed to teach people about the risks of prompt injection attacks on large language models (LLMs), until recently included an unintended expert level: a publicly accessible analytics dashboard that provided access to the prompts players submitted and related metrics.…

analytics attacks chatbot dashboard educational expert game language language models large language large language models llms people privacy prompt prompt injection prompt injection attacks prompts risks security