Skyflow Interceptor Adds Agent Tool Call Protection in Amazon Bedrock AgentCore Gateway
The Skyflow interceptor for Amazon Bedrock AgentCore Gateway allows organizations to transparently de-identify sensitive data in agent tool calls, results, and documents without requiring changes to agent or tool code. Developers can now adopt this open source solution to centralize data protection across AI workflows.
What changed?
Amazon Web Services and Skyflow have released an open source Lambda interceptor that integrates with the Amazon Bedrock AgentCore Gateway. Added as a gateway interceptor, it de-identifies sensitive data in agent tool arguments, tool results, and documents before they are handled by external tools or AI models. The interceptor can operate in three protection modes: Shield the tool, Shield the model, and Shield the document. Sensitive entities such as names and card numbers are replaced with tokens, which can later be re-identified for authorized users. The interceptor is centrally configured, requires no agent or tool code change, and is designed to fail-closed if inspection or de-identification cannot be completed.

Why does it matter to an everyday developer?
AI agent workflows increasingly move data between models, tools, and document stores, creating new risks for exposure of sensitive information. Traditional controls often only protect what enters or leaves the model. The Skyflow interceptor fills a control gap by enforcing consistent data protection for every tool call and result, regardless of a tool's vendor or its location. Developers and teams who work with LLM agents, third-party APIs, or sensitive customer data can now centrally enforce compliance and privacy policies through configuration, not code patches. This reduces the engineering burden and risk of accidental data leakage.
What can the developer do now?
Developers building AI agents with Amazon Bedrock AgentCore Gateway can deploy the Skyflow interceptor from the AWS Marketplace and configure it as an AWS Lambda function for their gateway. Policies can be defined in AWS Systems Manager Parameter Store, specifying which tools and data require de-identification and what type of tokenization to apply. The solution logs decisions to CloudWatch and never requires redeployment for policy changes. Developers can experiment by applying different modes to various tools, such as web search, document stores, or third-party APIs, and validate that no sensitive data flows outside their intended boundaries. No modification is needed to existing agent or tool applications.
