Amazon Web Services (AWS) has announced a new capability called Guardrails for Amazon Bedrock, which allows developers to customize safeguards for their AI chatbots and conversational agents. The goal is to promote safer, more responsible interactions between end users and generative AI applications.
The new guardrails integrate with Amazon Bedrock, AWS’s platform for building chatbots and voice assistants using large language models. Developers can set denied topics to restrict certain types of content, configure content filtering across categories like hate speech and violence, and redact personally identifiable information from chatbot responses.
According to AWS Vice President Antje Barth, the company aims to advance generative AI “in a responsible, people-centric way.” Guardrails help codify safety and privacy controls based on a developer’s specific use cases and responsible AI policies. Barth said this allows innovating safely while managing user experiences.
The guardrails apply across all large language models available through Amazon Bedrock, including fine-tuned models and third-party foundations like Anthropic’s Claude and AI21’s Jurassic. Developers can standardize controls consistently across multiple chatbots and assistants.
The denied topics feature lets developers define undesirable topics through natural language descriptions. For example, a banking chatbot could be set to avoid giving investment recommendations. Content filters go further by blocking outputs at adjustable thresholds for categories like sexual, violent, hateful or insulting content.
Amazon also plans to add personably identifiable information (PII) redaction soon. This will automatically remove or block sensitive user inputs like names, emails or phone numbers.
The guardrails integrate with Amazon CloudWatch for monitoring. Developers can analyze inputs or responses that violate defined policies to further improve systems.
The Guardrails for Amazon Bedrock capability remains in limited preview for now. Interested developers can request access through their AWS sales representative or support contacts. The preview supports major AI foundations like Claude, Llama and Titan Text, with custom models to follow.
By providing granular controls over safeguards and policy enforcement, Amazon aims to give developers more flexibility but also more accountability around responsible AI development using chatbots and conversational AI.
Initial image from aws.amazon.com