A credential-theft campaign that began on 10 November turned stolen GitHub access into a sweep of CI secrets across tens of thousands of repositories. Aikido Security, which tracks it as wfx-loot, says 4,837 compromised accounts opened probe pull requests in 61,764 repositories, and that the payload also plants cloud access built to survive credential rotation.

What Aikido found

According to Aikido’s analysis, published on 12 November, the threat actor used stolen GitHub tokens from about 12:45 UTC on 10 November. For each repository a token could write to, the actor first opened a pull request titled “docs: dependency audit notes” from a wfx-docs branch to confirm write access, then pushed the payload through one of three delivery routes: an injected step in an existing workflow, a postinstall hook in package.json, or a new workflow on a short-lived branch. Aikido says the actor reverted the change within at most one minute and deleted the workflow run along with its logs.

Aikido counts at least 1,835 repositories that ran the payload and had credentials exfiltrated, and says the number may understate impact because run logs were deleted. Of the probe pull requests, 10,431 went into repositories the account does not own, reached through collaborator or organisation access, and Aikido says 92% of them are still open, which suggests most owners have not noticed. At least 50 of the compromised users publish packages for ecosystems including npm, Go, Packagist, PyPI or crates. Aikido has identified the payload in one Packagist package and two tags of one Go module so far, and says the actor appears to be targeting CI secrets rather than spreading through the package supply chain.

Advertisement

CyberTech Your brand belongs here. Reach the decision-makers who read CyberTech every day. Premium placements across the site and newsletter. Advertise with us

Aikido’s guidance for affected teams is blunt: “If the payload ran in your repository, assume every secret and cloud role available to its workflows is compromised.” The write-up is by Oliver Smith of Aikido Security; the post does not give his title.

Why rotation alone does not close it

The stealer collects GitHub Actions secrets and variables, the workflow token, OIDC and runtime tokens, cloud credential files and AI provider keys. The part that should change incident response plans is what it does next. Aikido describes three persistence actions designed to outlast credential rotation:

  • Forged status checks. Early in the run, the payload posts success statuses using the names of the repository’s required checks. Aikido notes that branch protection matches on the context name alone, so the forged checks clear merge gates.
  • AWS trust policy rewrite. The payload sets the trust policy of an IAM role it chained into so that any AWS account can assume it.
  • Azure federated credentials. With a Microsoft Graph token, it adds federated identity credentials named wfx-ci to Azure AD applications, pointing back to the compromised repository, so workflows there can authenticate to Azure without a stored secret even after every token is rotated.

Aikido says the payload also mints OIDC tokens for AWS, GCP and Azure, creates long-lived IAM access keys where it can, impersonates GCP service accounts, and replays tokens against HashiCorp Vault and Kubernetes API servers it finds in kubeconfig files. On self-hosted runners it also reads instance metadata credentials.

How it connects to GhostAction

Aikido found that accounts hit by the October GhostAction wave were significantly more likely than baseline to be hit by wfx-loot, and that in more than 100 repositories wfx-loot was injected into GhostAction’s own workflow files. It still assesses the two as distinct campaigns, likely run by different actors, because the infrastructure, tooling and tradecraft differ, and it notes that stolen credentials often end up with several actors. We covered the earlier wave when GhostAction swept 345 repositories through two maintainer accounts. The overlap means an account that was cleaned up after GhostAction without a full token revocation may have been exposed again.

Newsletter

Get the week's best tech coverage.

Free. Read by thousands of HR, tech, and business leaders.

What it means for the security leader

In our reading, this campaign moves CI compromise from a secrets problem to an identity problem. Most playbooks for a leaked workflow secret end at rotation. Here, the attacker’s durable access lives in cloud trust relationships and federated credentials that rotation never touches, and in merge gates that accept a status from anyone who can name the check. The detection evidence is also thin by design: Aikido says a missing run number in a workflow’s history is often the only trace, and that the absence of wfx-loot artifacts or cache entries does not mean data was not taken.

The exposure also runs through people the organisation does not employ. More than 10,000 probe pull requests landed in repositories through collaborator and organisation access, so a contractor’s or open-source contributor’s stolen token can reach a company’s private repositories. The developer machines and AI coding tools that hold those tokens are part of the same surface, a point that came up when Microsoft detailed a tampered Claude Code extension earlier this month.

What defenders should do now

  • Search for the probe. Look for a wfx-docs branch, a pull request titled “docs: dependency audit notes”, short-lived branches named chore/ plus eight random characters, and the commit titles Aikido lists, in every repository you can write to, from 10 November on. Aikido says that if the probe ran, the user is compromised even if the workflow stage did not run.
  • Check for gaps in workflow run numbers around any of those commits, since deleted runs are the most reliable sign the payload executed.
  • Revoke and reissue every secret, token and cloud role available to workflows in an affected repository, then revoke the compromised user’s GitHub tokens.
  • Audit cloud trust after rotation. Review IAM role trust policies for wildcard principals, list new IAM user access keys, and check Azure applications for unexpected federated identity credentials, including any named wfx-ci.
  • Tie required checks to a specific app where your platform allows it, so a status posted by a workflow cannot satisfy a merge gate on its name alone.
  • Block egress from runners to the network indicators Aikido published, and remove the affected Packagist and Go package versions it lists.

Aikido also notes dedicated configuration in the payload for private repositories, which suggests deployment beyond the public repositories it can see. Teams that only checked public code should run the same searches across private organisations.

Source: Aikido Security: wfx-loot campaign pushes CI secret stealer to tens of thousands of repositories, 12 November 2026