LINK-V
LINK-V · Article

If Your Build Is Compromised, What Can It Reach?

November 15, 2026

Compare what a routine test and a production deployment can read or change. Find out who can edit or approve each job and who can withdraw its access.


If Your Build Is Compromised, What Can It Reach?

If malicious code runs in an automated build, it can attempt the actions that job is allowed to take. To see what is exposed, ask your team or supplier to show the effective permissions of one routine test job and one production deployment, with all credential values hidden. What can each job read or change, who can change the job itself, and who can withdraw its access?

What did the wfx-loot investigation show?

In its November 12 investigation, Aikido reported that an attacker used stolen GitHub credentials to check which repositories the accounts could write to. In a subset of repositories, malicious code ran and stole GitHub Actions secrets and cloud credentials available to the affected jobs. An access probe showed that the compromised account could write to a repository. Evidence of workflow execution was a separate finding.

A test or build job may run code your team did not write. The UK National Cyber Security Centre’s guidance on software dependencies explains how automated builds may retrieve and run third-party code. The access decision is the same whether unexpected code arrives through a repository change or a dependency. The permissions a job receives determine what that code can attempt.

What can a test job reach compared with a release?

Ask your operator to trace two real runs. For a web shop, the useful comparison might be a routine checkout test and a deployment that updates the checkout customers use.

Routine test job

Show where a proposed change runs and what it can read. For example, it might read source code and use sample data to test checkout. If testing can run without access to live orders or the power to release software, keep those permissions out of this job.

Production deployment job

Show which live service the job can change, what it can publish, and what has to happen before it can act. Check what it can reach through connected services as well as the initial credential it receives.

For each job, ask when it receives access, which role grants it and which other systems trust that role. Ask what happens if the job requests a permission it should lack. A configuration review or a controlled demonstration can show the boundary without attempting changes on the live service.

Who can change or approve a release?

Find out who can edit the instructions for the deployment, who approves a release, and who can change or bypass that approval. Where the platform supports it, keep approval under the control of someone who did not initiate the release and limit which job is allowed to act after approval.

GitHub’s guidance for using Actions securely describes setting minimum permissions, requiring review before jobs gain access to protected environment secrets, and using federated identity with supported cloud providers. With federation, the cloud can issue a short-lived credential to an authorized job. Check which job can request that credential and narrow the role to the service and actions the release actually needs. A credential with a short lifetime can still be used while it is valid.

Keep the evidence and name who can stop access

Ask for a short record of both jobs. Write down what each may read and change, how long its access lasts, who can edit or approve it, where you can find run and approval records, and who is authorized to withdraw each permission. Ask whether the same person can withdraw access in both the build platform and the cloud account. If a supplier controls one of them, name a contact and find out how to reach them.

Start by separating routine test access from production where the test permits it. Compare a narrower deployment role and an independent approval step where supported. Agree how to check any changes in stages so ordinary releases still work. If you are taking over a project and cannot yet run a build, our guide to handing unfinished software to a new developer covers what to establish first.

If you suspect the malicious workflow actually ran in your repository, involve your security operator and consult the original investigation. This access review is for understanding what the jobs can reach.

Review the infrastructure behind your releases

We work on custom software infrastructure. Bring the test and release jobs you want to review. We can discuss how to separate their access and who would approve releases and withdraw access.

Discuss your deployment setup

Questions about build and deployment access

Can a test job deploy to production?

It can if its permissions let it change the live service directly or trigger another job that can. Ask your operator to check both routes. Keep production access out of routine tests when they can run without it.

Can a short-lived credential still be abused?

Yes. Code may use it within its permissions while it is valid. Limit what it can change and check which jobs can request a fresh credential.

What can a supplier show without exposing credentials?

Ask for the job name, the role it uses, the permissions in effect and a record of an approved run with sensitive values hidden. Have them name the person who can withdraw the access.

LINK-V LINK-V