Arthur Shield by Arthur is a firewall designed for Large Language Models (LLMs) that sits between the application layer and the deployment layer. It validates user prompts and model responses on two endpoints to protect against major LLM threats in production environments. Arthur Shield enables safe deployment of LLMs by protecting sensitive data from leakage, blocking non-aligned responses, preventing prompt injections, detecting hallucinations, and mitigating various risks associated with LLMs. It aims to address issues such as personal identifiable information (PII) exposure, generation of toxic or problematic language, and maintaining the integrity of LLM applications.
