← All articles

GitHub secret stealer hit 1,806 repositories

GitHub secret stealer hit 1,806 repositories

A GitHub secret stealer was planted in at least 1,806 repositories after an attacker used stolen credentials from 4,837 GitHub accounts, according to Aikido Security. The malware targets continuous integration (CI), the automated system that builds and tests software, and collects secrets and cloud credentials available to repository workflows.

The campaign began Tuesday. Each initial test used a pull request, a proposed change submitted for review, titled “docs: dependency audit notes” from a branch called wfx-docs. These tests reached 61,764 repositories to check where the compromised accounts could write. The stealer was then added to some of them. A separate analysis by SafeDep counted 8,093 public repositories that received malicious commits between 12:46 UTC on Tuesday and 04:36 UTC today. That figure includes repositories where the stealer did not run, and covers only default branches of public repositories, so the real total is higher.

In 7,309 repositories, the code was inserted into an existing workflow. In 771, it arrived as an npm postinstall hook, which runs on a developer’s computer during npm install. The attacker generally removed the changes within minutes. Where the stealer ran, logs and often the workflow run itself were deleted. Aikido says every secret and cloud role that the workflow could access should therefore be treated as compromised.

More pull requests with the same title appeared early today, after Aikido’s own GitHub search ended on Wednesday. The Hacker News checked three opened between 05:04 and 05:38 UTC, all from wfx-docs branches, although it was not clear whether the stealer followed. Aikido says any account that opened one of the attacker’s test pull requests is compromised, even when the stealer never ran. The analysis does not explain how the attacker obtained the credentials.

The campaign also abused access beyond repositories owned by the compromised accounts. Of the test pull requests, 10,431 were submitted through collaborator or organisation access, and 92% remained open when checked. In most other repositories, nothing was pushed after the test, or repository settings and account permissions stopped the stealer from running. The published figures cover only public repositories visible to the researchers.

Repository owners should look from Tuesday onward for a wfx-docs branch, short-lived branches named chore/ followed by eight lowercase letters or numbers, and branches called ci-check- followed by four numbers. Other warning signs include commits titled “ci: refresh build cache”, which adds about 1,500 lines to an existing workflow; “chore: add ci helper”, which adds a .ci-xxxxxx.sh script and an npm postinstall hook; “chore: add dependency audit workflow”, which adds a .github/workflows/build-audit-xxxxxx.yml file on a chore/ branch; “chore: initial commit” before a new workflow file in an empty repository; and “ci: restore” or “ci: cleanup”.

The stealer’s SHA-256 value, a fingerprint used to identify a file, is 89695abb392d5cfd4972672961e644bb46baa98aa1ed7173c2a2816c762735cf. SafeDep found that data was sent by HTTP POST, a web request used to upload information, to 194.150.166[.]244:8080/exfil, 18.143.168[.]96:8080/exfil or 153.75.82[.]159:8080, and also to a Telegram chat. In a sample of 200 payload commits, about one fifth used Telegram only. A missing run number on the attacker’s commit is a more reliable sign that the stealer ran than the absence of wfx-loot artifacts, wfx- cache entries or WFX-BEGIN step summaries.

The malware is designed to gather credentials from every source available on a CI runner. It can take GitHub Actions secrets and variables, the workflow’s GITHUB_TOKEN, OpenID Connect (OIDC) tokens used to prove a workload’s identity, Actions runtime tokens, AI service keys and secrets held in running processes. It reads credential files for AWS, Azure, Google Cloud Platform, Kubernetes, npm, PyPI, SSH and other tools. It also copies the repository, its last 100 commits and recent workflow artifacts. On self-hosted runners, it can collect cloud credentials from instance metadata, along with Docker container state and systemd units.

The code includes settings for private repositories, which Aikido says suggests that it may be running in them too. It can exchange GitHub OIDC tokens for AWS, Google Cloud and Azure sessions. On AWS, it can create long-lived access keys for Identity and Access Management users. Some actions are designed to survive normal credential rotation: the stealer can change an AWS role’s trust policy so any AWS account can use the role, and can add Azure federated identity credentials with the prefix wfx-ci- to applications. A compromised workflow could then sign in to Azure without a stored secret even after tokens are rotated.

Researchers recommend checking Entra ID for unrecognised federated credentials named wfx-ci-*, AWS CloudTrail for AssumeRoleWithWebIdentity or AssumeRole sessions named wfx-* followed by CreateAccessKey or UpdateAssumeRolePolicy, Google Cloud logs for GenerateAccessToken and SignJwt calls from the GitHub workload identity pool, and Vault for JWT logins from the GitHub OIDC issuer.

The campaign reached at least one Packagist package and one Go module through automated publishing triggers. At least 50 compromised users appear to publish packages for npm, Go, Packagist, PyPI or crates. Aikido says the attacker appears focused on CI secrets rather than spreading through packages.

Packagist package digitalelvis/neuronai-studio version 3.14.3 is malicious. Its listing showed that the version was published on Wednesday and was still available early today under a warning that package versions had been flagged as malware. A normal composer install does not run the stealer, but npm install inside the package’s vendor directory does. Users should block version 3.14.3 and pin version 3.14.2 or earlier, while treating any machine that installed or opened 3.14.3 as compromised.

Go module github.com/onuragtas/docker-env versions v1.0.278 and v1.0.280 contain the stealer only in a GitHub Actions workflow file. The Go code and release binaries are unchanged, and Go builds do not run workflow files, so module users may not have been affected. The company nevertheless recommends removing those versions. Another Go module, github.com/vkuznet/gocurl, contains the stealer in pseudo-version v0.0.0-20261111095532-8d5aad0cb0da, inside a workflow that has never run. Users should block that pseudo-version.

The stealer also reached 28 WebdriverIO repositories through commits attributed to the wdio-bot account. No package release was affected, but the bot’s token should be considered stolen. The accounts used in this campaign overlap with those targeted in GhostAction’s October wave, when stolen GitHub accounts pushed secret-stealing workflows into repositories. Accounts hit in October were about 1,000 times more likely than other active accounts to be used this week.

Also available in: DeutschItalianoShqip