Announcing Guardrails for Amazon Bedrock now available with new safety filters and privacy controls

Guardrails for Amazon Bedrock has been officially released, offering a comprehensive solution for implementing safeguards in generative AI applications. These guardrails are customizable to specific use cases and AI policies, allowing users to create multiple guardrails and apply them across various foundation models (FMs). With this release, end-user experiences can be improved, and safety controls can be standardized across generative AI applications.

Guardrails for Amazon Bedrock enhances safety protection beyond the native capabilities of foundation models, providing industry-leading safeguards. It can block up to 85% more harmful content compared to the native protection offered by some foundation models on Amazon Bedrock. Notably, it is the only responsible AI capability from a major cloud provider that enables users to build and customize safety and privacy protections for their generative AI applications in a single solution. It works with all large language models (LLMs) in Amazon Bedrock, including fine-tuned models.

Aha!, a software company, leverages Amazon Bedrock to power many of its generative AI capabilities. By utilizing Amazon Bedrock's responsible AI features, Aha! maintains control over its information through data protection and privacy policies, while also leveraging Guardrails for Bedrock to block harmful content. Aha! plans to further utilize AWS technology to assist product development teams in prioritizing their next steps with confidence.

Guardrails for Amazon Bedrock sits between the application and the model, automatically evaluating inputs and outputs to detect and prevent content that falls into restricted categories. The recent update introduces new features such as sensitive information filters, word filters, and additional safety categories like Misconduct and Prompt Attack.

To configure Guardrails for Amazon Bedrock, users can access the AWS Management Console for Amazon Bedrock, where they can create guardrails and configure new capabilities. Sensitive information filters can be used to detect and redact sensitive information, while word filters can block inputs containing profane or custom words. Users can customize messages displayed when inputs or model responses are blocked.

Guardrails for Amazon Bedrock is now available in US East (N. Virginia) and US West (Oregon) Regions. Users can visit the Amazon Bedrock pricing page for pricing information and the Guardrails for Amazon Bedrock web page to get started with this feature. Additionally, the community.aws website offers deep-dive technical content and insights into how Builder communities are utilizing Amazon Bedrock in their solutions.

Source :- https://aws.amazon.com/it/blogs/aws/guardrails-for-amazon-bedrock-now-available-with-new-safety-filters-and-privacy-controls/