Putting access controls around GitHub Copilot: audit trails for AI coding agents (on AWS)

Why audit trails matter for AI coding agents

When an AI assistant like GitHub Copilot writes code directly into a repository, every line it produces becomes part of the production pipeline. If a secret is inadvertently committed, or a risky change is pushed without human review, the impact can spread across services in minutes. An audit trail that records who, what, and when the AI acted is the only reliable way to detect abuse, satisfy compliance reviews, and roll back unintended changes.

How teams currently give Copilot access on AWS

Most organizations integrate Copilot with their AWS‑hosted development environment by provisioning a static GitHub token or an IAM role that the Copilot service can assume. The token lives in a CI/CD secret store, is checked out by the AI runtime, and is used for every push, pull, or branch creation. Engineers share the same credential, and the AI agent never authenticates as an individual user. The result is a single point of trust that grants unlimited write access to every repository the token can reach.

The hidden risk

Because the token is static, any compromise, whether through a leaked secret, a malicious fork, or a mis‑behaving model output, immediately grants full write privileges. There is no built‑in visibility into which Copilot request caused a change, and no way to enforce a review before code lands. The system logs that AWS or GitHub emit are coarse, showing only that the token was used, not which AI prompt triggered the operation.

The missing control – audit trails without a gateway

What teams often try to add is a separate logging service that scrapes GitHub audit events after the fact. That approach still leaves the request path untouched: the AI agent talks straight to GitHub, the token is presented unchanged, and any dangerous command is executed before the logging service can react. In this configuration, audit trails are retroactive, incomplete, and cannot prevent a harmful push.

Introducing hoop.dev as the access gateway for Copilot

hoop.dev is a Layer 7 gateway that sits between the AI runtime and the GitHub API. The gateway runs a network‑resident agent inside the same VPC where the development environment lives. Identity is handled by OIDC; engineers authenticate with their corporate IdP, and the resulting token is presented to hoop.dev, which then decides whether the request may proceed.

Because hoop.dev is the only point where traffic passes, it can enforce the controls that matter:

• Audit trails: hoop.dev records each Copilot request, the user identity that initiated it, the exact API call, and the response payload. The recorded session can be replayed later for forensic analysis.
• Just‑in‑time approval: before a push that modifies a protected branch, hoop.dev can pause the request and require an authorized reviewer to approve it.
• Inline masking: if a response contains a secret token, hoop.dev can redact the value before it reaches the AI agent, preventing accidental exposure.
• Session replay: every interaction is stored in a replayable log, allowing teams to step through an AI‑generated change line by line.

All of these outcomes exist only because hoop.dev occupies the data path. The setup stage, defining OIDC client, provisioning a service account, and configuring the GitHub connector, decides who may start a request, but it does not enforce any policy on its own. hoop.dev is the enforcement point that turns a raw token into a controlled, auditable channel.

Getting started

To protect your Copilot workflow, deploy hoop.dev using the Docker Compose quick‑start, configure the GitHub connector with the static token you already use, and enable the audit‑trail feature in the UI. The gateway will automatically intercept all Copilot API calls, apply the policies you define, and store the logs for later review. Detailed steps are covered in the getting‑started guide, and the broader feature set is described on the learn page.

Explore the open‑source code, contribute improvements, or fork the repository at GitHub.

FAQ

Do I need to change my Copilot configuration?

No. hoop.dev works as a transparent proxy; you simply point the Copilot client at the gateway endpoint and keep using the same GitHub token.

Does hoop.dev store my GitHub credentials?

No. The gateway holds the credential in memory for the duration of each session and never exposes it to the AI runtime.

Can I view past sessions?

Yes. hoop.dev records each session and provides a UI for replaying the full request‑response flow, giving you a complete audit trail for any period you need.