Tag: prompt injection

AI application security is not solved by a system prompt. Protect the application with least-privilege tools, permission-aware retrieval, input and output controls, evaluation, and traces that make failures diagnosable.