How autonomous agents processes data

Autonomous agents uses Reltio-managed AI services to process requests. The data processing behavior depends on the service involved:

• Agent operations: Autonomous agents sends requests to Anthropic Claude models through Amazon Bedrock for agent conversations, long-running tasks, specialized agents, custom instructions, and custom agents.
• Guardrails: Autonomous agents use Anthropic Claude models through Amazon Bedrock to evaluate prompts and responses against the autonomous agents usage policy.
• Web search: When enabled, autonomous agents sends web search requests to Google Cloud Vertex AI or Parallel.

Web search is optional and disabled by default. For information about the models used by each capability, see Large language models and AI services in autonomous agents.

Where AI data processing takes place

Autonomous agents uses Amazon Bedrock in AWS US regions for Reltio-managed LLM processing. The processing environment includes:

• Amazon Bedrock: Autonomous agents use Amazon Bedrock cross-region inference, which allows AWS to route a request to any AWS US region to balance capacity. Requests are not routed outside the United States.
• Customer Dataplane: When Customer Dataplane is enabled, autonomous agents guardrail checks run in your organization's AWS account.
• Optional web search services: Web search requests are processed by Google Cloud Vertex AI or Parallel when web search is enabled.

How Amazon Bedrock handles your data

Reltio-managed LLM processing through Amazon Bedrock follows these data-handling practices:

• Amazon Bedrock does not store prompts or model responses.
• Amazon Bedrock does not share prompts or responses with Anthropic.
• Amazon Bedrock does not use prompts or responses to train models.

Retention of that data is described in Autonomous agents security, audit, and compliance architecture

For more information, see Data protection in Amazon Bedrock.

Data shared with web search services

When web search is enabled, autonomous agents may send the following information to Google Cloud Vertex AI or Parallel:

• Search queries generated during an agent task.
• Webpage addresses that the agent needs to read.
• Your data included in a search query or webpage-reading request.

Because these requests may include your data, web search introduces an additional external data-processing path beyond Amazon Bedrock. If your organization does not allow data to be sent to these services, keep web search disabled.

Model training and data usage

Reltio does not use your data to develop or customize the LLMs used by autonomous agents. Specifically, Reltio does not:

• Train LLMs using customer data.
• Fine-tune LLMs using customer data.
• Create embeddings or vector indexes from customer data as part of the described autonomous agents LLM processing.