-
Wiz found more than 13,500 planted branches across more than 850 hijacked accounts. The attacks worked in private repositories.
-
GitHub’s safety hold stopped every public attack Wiz checked, but it doesn’t cover private repositories or GitHub Enterprise Server.
-
Deleting the branch doesn’t undo the damage. Teams must change every secret and cancel the stolen token.
An automated campaign is stealing passwords and keys from private GitHub projects. Security firm Wiz has tracked it since September 15.
It calls the campaign Gheist, after a tag the attackers’ tool sends with its requests. The attackers use stolen access tokens. A token works like a password for software, so the attackers simply act as the victim.
How the Attack Works
If a token has write access to a repository, it can also read every secret stored there. GitHub’s own documentation says so. For each repository a token can reach, the tool creates a branch.
Its name starts with ci-sync- and ends with six random characters. Next, the tool adds one file, sync-check.yml, with the message “ci: sync config.” The commit carries the token owner’s name but has no signature. The push starts the workflow.
The workflow prints the job’s environment as well as any secrets that it can access. The text is scrambled into a code called base32. Secrets in GitHub logs only get obscured when they exactly match. The scrambled text slips past that filter and prints in full.
One version also runs a job for each deployment environment. As a result, secrets tied to those environments print too. Wiz says these are often production keys.
Then the tool downloads the log through GitHub’s own system. Seconds later, it deletes the run. Because the data never leaves GitHub, no attacker server or outside link exists to block.
Where GitHub’s Defense Falls Short
Since July 28, GitHub has paused workflow runs that look harmful. A collaborator with write access must approve them first. Every public Gheist run that Wiz checked sat waiting, so its payload never ran.
However, GitHub says the hold covers only public repositories on github.com. GitHub Enterprise Server does not have it. In private repositories, audit logs show runs finishing, and then the gheist tag deleting them. Meanwhile, AWS logs show company workflows running on the planted branches and taking on cloud roles. That proves the code ran without approval.
Separately, Anthropic has launched an AI security tool designed to find software vulnerabilities, reflecting broader efforts to identify security weaknesses before attackers exploit them.
In large companies, the tool cycled through tokens from several users and automation accounts within minutes. It also copied private repositories before planting its file. Hours later, someone used the company’s cloud credentials to list AWS Secrets Manager entries. Wiz does not say who. It also does not say how many organizations lost secrets.
Around October 19, one version added a second job for self-hosted runners. These are machines that companies run themselves to handle GitHub jobs. The job hunts for 31 kinds of tokens and keys. Still, Wiz calls it buggy and reports no successful runs.
As of October 24, the campaign was still running and changing. About two-thirds of the accounts involved are gone from GitHub. Wiz reported the activity to GitHub and believes the disappearances likely reflect a response.
The Attackers Steal Tokens Using Malware
Wiz traced some of the stolen tokens to logs from password-stealing malware. Others came from earlier supply-chain attacks, such as the March breach of the Trivy scanner. Other criminals use the same accounts, too.
On October 6, 391 accounts renamed about 1,461 repositories to look like cracked software and spread malware. Some victims also appear to carry PolinRider, malware that researchers link to North Korea.
Gheist is not the same as GhostAction, a separate campaign with different workflow files. According to GitGuardian, a new GhostAction wave ran in public repositories until GitHub began blocking it on October 9.
Recommended Response
Check Your Repositories
Wiz suggests four checks:
- Search for branches named ci-sync- plus six characters. The only change should be a new sync-check.yml file with the message “ci: sync config.”
- Review GitHub audit logs for runs on those branches that vanished in under a minute. Also look for the tag gheist.
- If your workflows use OIDC, search cloud logs for sign-ins tied to a ci-sync- branch. OIDC gives workflows short-lived passes instead of stored keys. This works even without GitHub logs.
- On self-hosted runners, look for the files .gdmp_pat, .gdmp_patg, and .gdmp_sweep.
What to Do If You Find One
Change every repository, organization and environment secret the repository could use. Cancel the token that pushed the commits, and scan its owner’s devices for malware. Treat any self-hosted machine that ran the branch, and its cloud role, as compromised. Do not approve held runs or run code from these branches.
Next, make stolen tokens less useful. Require reviewers on environments that hold production secrets. GitHub says a job cannot read those secrets until a reviewer approves it. Limit cloud trust rules to protected branches or environments.
Use GitHub Apps or short-lived tokens instead of personal access tokens. Finally, run self-hosted runners as short-lived machines.