A GitHub CI secret stealer swept through the platform this week, and the scale is hard to overstate. Starting on Tuesday 10 November 2026, an attacker used stolen credentials for 4,837 GitHub accounts to push malicious code into tens of thousands of repositories. Aikido Security, which first documented the campaign and named it wfx-loot, says the payload ran in at least 1,806 repositories. A separate count by SafeDep put repositories receiving malicious commits at 8,093. If you keep code on GitHub, this is worth five minutes of your time.
The attack did not exploit a flaw in GitHub itself, so no patch is coming that fixes it for everyone. The attacker simply logged in with stolen credentials and used the access those accounts already had. Every account that opened one of the attacker’s test pull requests is compromised, Aikido says, even if the malicious code never ran in that repository.
How the GitHub CI Secret Stealer Works
The campaign followed a simple three-step pattern. First, the attacker opened a probe pull request titled “docs: dependency audit notes” from a branch called wfx-docs. That pull request was only a test: if it went through, the attacker knew the stolen account had write access. Aikido counted 61,764 repositories that received one between 12:45 UTC on 10 November and 15:00 UTC on 11 November, with more still appearing on Thursday 12 November.
Second, where the account could write, the attacker pushed the payload. It arrived in one of three forms: a commit titled “ci: refresh build cache” that added about 1,500 lines to an existing workflow file, a commit titled “chore: add ci helper” that added a shell script plus a postinstall hook in package.json, so the code also runs on any developer laptop that installs the package, or a commit titled “chore: add dependency audit workflow” that created a new workflow file on a short-lived branch.
Third, the attacker cleaned up. Usually the malicious change was reverted within a minute, and the workflow run and its logs were deleted. That is what makes this campaign so awkward to investigate: the evidence deletes itself. According to Aikido, a missing run number in a workflow’s history is often the only trace left behind, and a more reliable sign the payload ran than searching for leftover files.

What the Attackers Were After
The payload is a single shell script of about 1,500 lines, and it is thorough. It collects GitHub Actions secrets and variables, the workflow GITHUB_TOKEN, OIDC tokens and Actions runtime tokens, and sweeps the runner’s environment for anything matching words like token, secret, password, AWS, Azure, GCP, OpenAI, Anthropic and Stripe. It reads credential files for AWS, Azure, Google Cloud, Docker, Kubernetes, npm, PyPI, Git, SSH and Terraform, plus every .env, .pem and kubeconfig file under the home directory, and hunts specifically for AI provider keys from OpenAI, Anthropic, Gemini, Hugging Face, Cursor, Copilot and Codex.
Then it tries to turn those credentials into deeper access. It mints GitHub OIDC tokens and exchanges them for cloud sessions on AWS, GCP and Azure. On AWS it chains into other IAM roles and creates long-lived access keys. On Azure it lists subscriptions and Key Vaults. The campaign also planted persistence tricks designed to survive a password reset: on AWS it rewrote a role’s trust policy to open it to any AWS account, and on Azure it added federated identity credentials named wfx-ci- plus random characters to existing Entra ID applications, pointing back at the compromised repository. That last trick means a workflow there can keep signing in to Azure even after every token has been rotated, which is why simply changing passwords is not enough here.
The stolen data went by HTTP POST to two callback servers, at 194.150.166.244:8080 and 18.143.168.96:8080, both on the /exfil path. SafeDep also found evidence of exfiltration through a Telegram channel.
The GhostAction Connection
If this campaign feels familiar, there is a reason. Aikido found heavy overlap between the accounts used here and those hit by the GhostAction campaign in October 2026, when stolen GitHub accounts pushed secret-stealing workflows into repositories in a very similar way. In more than 100 repositories, the wfx-loot payload was injected into GhostAction’s own workflow files.
Aikido is clear, though, that the two are probably different actors: the infrastructure, tooling and tradecraft differ, and stolen credentials often end up with several threat actors at once. Researchers also spotted Indonesian-language comments in newer versions of the malware, never seen in GhostAction activity. The lesson is the same either way: once developer credentials leak, they tend to get reused again and again.
How to Tell If Your Repositories Were Hit
Check for these signs in repositories you can write to, from 10 November 2026 onwards: a wfx-docs branch, a pull request titled “docs: dependency audit notes”, commits titled “ci: refresh build cache”, “chore: add ci helper” or “chore: add dependency audit workflow”, or short-lived branches named chore/ plus eight random characters. Aikido says 92 percent of the probe pull requests were still open when it published its analysis, which suggests most owners had not noticed them yet.
Also check your workflow run history for gaps: a missing run number on one of the attacker’s commits is the strongest sign the payload executed, because the attacker deleted the run and its logs. Check whether your projects depend on the two affected packages: version 3.14.3 of the Packagist package digitalelvis/neuronai-studio, and versions 1.0.278 and 1.0.280 of the Go module github.com/onuragtas/docker-env. Aikido advises treating any machine that installed the Packagist package as compromised. The Go module only carried the payload in a workflow file, so Go builds were not affected, but the tags should still be removed.

What to Do Now
If you find any of the signs above, assume every secret and cloud role available to that repository’s workflows is compromised and rotate them all: GitHub Actions secrets, cloud access keys, API tokens and AI provider keys. Then go further: on AWS, check IAM roles for trust policies open to any account and for access keys you did not create. On Azure, remove any federated identity credentials named wfx-ci- on your Entra ID applications, because rotating tokens alone will not close that door. Review audit logs from 10 November onwards, and revoke and reissue any GitHub tokens that may have been exposed.
For everyone else, this is a good week to tighten the basics. Turn on two-factor authentication for every GitHub account on your team, and consider hardware security keys for accounts with production access. Keep secrets out of repositories, and give workflows the minimum permissions they need. Self-hosted runners are a bigger target than GitHub-hosted ones, because the payload can also read instance metadata and Docker state from them. Finally, pin third-party GitHub Actions to specific commit SHAs instead of floating tags, so a compromised action cannot silently change underneath you.
For background on how supply chain incidents unfold, see our report on the DigiCert data breach and our coverage of the Fireworks AI breach. Keeping your tooling patched matters too, which is why Microsoft’s November Patch Tuesday, which fixed 1,093 flaws including an exploited Outlook zero-day, deserves your attention this week.
What is the wfx-loot GitHub CI secret stealer?
It is a malicious shell script, about 1,500 lines long, that an attacker pushed into GitHub repositories using stolen account credentials from 10 November 2026. It steals GitHub Actions secrets, cloud credentials and AI provider keys from CI runners, then sends them to attacker-controlled servers. Aikido Security named the campaign wfx-loot after the name the malware uses for itself in its own logs.
How many GitHub accounts and repositories were affected?
Aikido counted 4,837 compromised accounts that opened probe pull requests in 61,764 repositories. The payload ran in at least 1,806 repositories according to Aikido, and SafeDep counted 8,093 repositories that received malicious commits. The numbers differ because the firms measured at different times and by different methods, and both say their counts may understate the real impact.
Was GitHub itself hacked?
No. The attacker did not exploit a vulnerability in GitHub Actions. The campaign used legitimately issued but stolen credentials, so the attacker logged in as a normal user and used the access those accounts already had. That is why the fix is credential hygiene and tighter permissions, not a patch from GitHub.
What data did the stealer collect?
It collected GitHub Actions secrets and variables, workflow tokens, OIDC tokens, cloud credentials for AWS, Azure and Google Cloud, credential files for Docker, Kubernetes, npm, PyPI, Git, SSH and Terraform, shell histories, and API keys for AI providers including OpenAI, Anthropic, Gemini, Hugging Face, Cursor and Copilot. It also searched for .env, .pem and kubeconfig files.
How do I know if my repository was hit?
Look for a wfx-docs branch, a pull request titled “docs: dependency audit notes”, commits titled “ci: refresh build cache”, “chore: add ci helper” or “chore: add dependency audit workflow”, or short-lived chore/ branches. A missing run number in your workflow history is the strongest sign the payload ran, because the attacker deleted the run and its logs.
I only received a probe pull request. Am I safe?
No. Aikido states clearly that any account that opened one of the attacker’s test pull requests is compromised, even if the stealer never ran. Change that account’s credentials, review its recent activity, and check every repository it can write to.
What should I do if my repository was affected?
Rotate every secret and cloud role the repository’s workflows could reach, then check for persistence tricks: AWS trust policies opened to any account, and federated identity credentials named wfx-ci- on Azure Entra ID applications. Revoke exposed GitHub tokens, remove the affected package versions, and review audit logs from 10 November 2026 onwards.
#GitHub #CyberSecurity #SupplyChainAttack #DevSecOps #CISecurity #SecretStealer
One Response