Overview
Severity: MEDIUM | Affected: Aegis AI Foundation | Category: tool
The Aegis AI Foundation has launched LLM-Guard v1.0, a new open-source security tool designed to protect applications built on top of large language models. Functioning as a highly configurable API gateway, LLM-Guard inspects both incoming prompts and outgoing responses in real-time to mitigate a range of threats. Key features include scanners for prompt injection, detection of data leakage, PII redaction, and a filter to prevent the model from generating toxic or harmful content. It is designed to be model-agnostic, allowing developers to integrate it with APIs from providers like OpenAI, Google, and Anthropic. The foundation states that LLM-Guard aims to provide a standardized, community-driven layer of defense, enabling developers to more safely deploy AI-powered applications without having to become experts in the nuances of adversarial AI. The project is available on GitHub under the MIT license.