How to audit what your AI agents are accessing
Learn how to audit AI agents with identity-based logging, LLM request tracking, data controls, tool authorization, and pre-request security policies.
Aperture is a tool that allows users to audit the data accessed by their AI agents. It provides identity tracking for every AI request, which includes the identity of the person or device making the request. Aperture groups related requests into sessions, making it easy to see the context of the data, tools, and costs associated with deeper work.
Aperture captures the full request-response lifecycle of every LLM interaction, including the full request body, full response body, HTTP headers (with sensitive values redacted), token counts by type, model name, duration, and tool use. This data is captured asynchronously, without slowing down AI work.
The retention period for captured data can be customized, with the option to reduce it to zero. Logs can be exported to S3-compatible storage for integration with SIEM systems to detect and respond to threats. Aperture also supports integration with third-party tools such as Cribl, Oso, Apollo Research, and Cerbos.
Access to logs is controlled through a grants system, which can be configured based on Tailscale identities. There are two roles currently available: user and admin. Administrator access logging records who viewed logs owned by another user. Aperture's policies ensure data is enforced before it leaves the network, allowing for guardrails to be set, personally identifiable information (PII) to be scrubbed, and requests to be blocked if they violate data policy.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
