AWS Re:Invent Daily Update Blog - 11/28
Today at AWS Re:Invent, the overarching theme was generative AI and AWS’s advances in that space. While not surprising, these updates can be challenging to absorb as they change the way many of us think about operating in the cloud security. As such, we wanted to identify a few updates that caught our attention.
Amazon DataZone is a data management service for cataloging, discovering, and governing data stored across AWS and other sources. While these activities are crucial for providing data context to users, too often data management either falls short or takes far too much time. As such, Amazon has announced generative AI capabilities within Amazon DataZone to automate aspects of data context.
When DataZone was launched, AI capabilities to automate the generation of table and column names of a data catalog were included. Today, Amazon announced that the generation of detailed descriptions of tables and schemas will be supported, in addition to suggested uses of the data. These descriptions can be generated then accepted, edited or rejected. For schema, the generation of the “Name” field is also supported. This new feature will further reduce the tedious manual work involved in maintaining a data catalog and will increase the data value to business users.
|Overwhelmed by AI and its implications around data management and security? Look no further than our eBook, Maximize Generative AI to Accelerate Your Business Securely. Additionally, feel free to reach out to us should you have any questions around generative AI or security in general.
As Amazon continues to offer more innovative products leveraging generative AI, the power that AI brings can be concerning for enterprises. As such, Amazon has announced Guardrails for Amazon Bedrock, which allows the implementation of safeguards for AI use cases in your AWS environment. These guardrails can be used to define denied topics and implement content filters to safeguard interactions between users and applications. These protections provide an extra layer of protection in addition to the protections built into AI models themselves.
Guardrails can be applied to all large language models (LLMs) within Amazon Bedrock. This allows for standardization of controls across all of your organization’s AI applications. Guardrails can include overarching topics (financial advice, legal topics, etc.) and content filters such as hate, insults, etc. Content filters can have filter strengths ranging from “None” to “High”. Additionally, PII redaction is currently in development to filter the inclusion of PII in AI responses or user input. Guardrails for Amazon Bedrock also integrates with CloudWatch so policy violations can be monitored.