Security7 min read
Securing language-model systems: prompt injection
If a model reads text from outside your organisation, that text is code arriving at the model. It is a class of vulnerability with no direct equivalent in classical software.
Read on