Skip to content
context Security

Skyflow Detect now protects sensitive data in AI agent tool calls in Amazon Bedrock AgentCore Gateway

only one source so far

Skyflow has integrated the Skyflow Detect tool into Amazon Bedrock AgentCore Gateway as an interceptor that de-identifies sensitive data before it is sent to external tools and restores it before the agent reads it. The solution is available in the AWS Marketplace, and the interceptor is open source on GitHub.

Skyflow, in collaboration with Amazon Web Services, has integrated the Skyflow Detect tool into the Amazon Bedrock AgentCore Gateway service. Skyflow Detect works as a request and response interceptor: before sending a call to a tool (MCP servers, SaaS APIs, document storage, web search), it removes sensitive data such as names, emails, phone numbers, card numbers, or dates of birth and replaces them with tokens; after the response returns, it restores them to their original values before the agent reads them. For documents, according to the company, the interceptor replaces the link to the original file with a link to a de-identified copy, so the agent never opens the original.

According to Skyflow, the solution is inserted directly into the gateway, so it works without any change to the agent's or tool's code and protects all agents and tools passing through that gateway uniformly. It offers three protection modes: "Shield the tool" (data hidden from the tool, visible to the agent and user), "Shield the model" (data hidden even from the model itself, restored only in the response), and "Shield the document" (replacing the signed URL with a link to a cleaned copy of the document).

As an example, the source cites a fictitious situation of a European bank with data residency obligations, where an agent resolving a payment dispute previously sent the client's name to an external search service; with the interceptor, the external provider receives only a reversible token instead of the name. The solution is available in the AWS Marketplace and the interceptor is open source on GitHub. The source text was only partially available, and the remaining technical details regarding the Skyflow Detect v2 version are not documented.

What changed

Why it matters

Companies deploying AI agents with access to external tools and APIs have so far had little control over what data the agent sends to third parties in the arguments of tool calls or what is returned back into the model's context. Central protection at the gateway level solves this problem without the need to modify each agent or tool individually, which is particularly relevant for regulated industries (finance, healthcare) with requirements for personal data protection and data residency.

Relevant practical impact

What this means

01

For a business

Companies running AI agents with access to external tools can centrally, at the gateway level, prevent sensitive customer data (name, email, card number) from leaking to third parties, without having to modify the code of individual agents or tools, which reduces the risk of compliance violations and data leaks.

Risks and compliance
What to decide Verify whether AI agents in the company that pass through AgentCore Gateway send sensitive data to external tools, and consider deploying a similar interceptor.
More business impacts →
AgentCore Gateway AI agents Amazon Bedrock AWS Data protection Skyflow

Check the original

Event sources

only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.

1
AWS Machine Learning Blog primary source · first detected Protect agent tool calls with Skyflow and Amazon Bedrock AgentCore Gateway