An attacker used a compromised GitHub account and leaked credentials for four weeks, copying customer contacts, contract terms, SQL editor text and encrypted API keys. PingCAP says the databases themselves were not reached.
Gheist campaign uses stolen GitHub tokens to dump Actions secrets
EditorialBy TrustList Editorial
Wiz Research tracked more than 13,500 planted ci-sync branches from over 850 compromised accounts; GitHub's approval hold stops the workflows in public repositories but not in private ones or GitHub Enterprise Server.
- DevOps
- Credential Management
- Application Security
- Cloud Security
- +2 more
About Gheist campaign uses stolen GitHub tokens to dump Actions secrets
Gheist campaign uses stolen GitHub tokens to dump Actions secrets
25 October 2026: A campaign Wiz Research calls Gheist has been using stolen GitHub tokens since 15 September to plant a workflow in every repository each token can reach and print the repository's secrets into the workflow log. As of 24 October, Wiz had recovered more than 13,500 planted branches from more than 850 compromised accounts, from personal projects to the private repositories of enterprise organisations. Roughly two thirds of the accounts are no longer available on GitHub, which Wiz reads as response activity after it reported the campaign.
Not yet independently verified. This rests on Wiz Research's own report of 25 October 2026. We have no statement from GitHub, and the account and branch counts are Wiz's figures as of 24 October. We will update this when it can be confirmed, and remove this note.
How the attack works
For each repository the tool creates a branch named ci-sync- plus six random characters and commits .github/workflows/sync-check.yml with the message "ci: sync config", unsigned and authored as the token's owner. The push triggers the workflow, which prints the environment and the whole secrets context encoded in base32. GitHub masks secrets in logs only when they appear verbatim, so the encoded values print in full. Where a repository has deployment environments, the tool runs a job per environment, so environment-scoped secrets, often production credentials, are dumped too. The tool then downloads the log through the GitHub API and deletes the run within seconds. Because the secrets leave through GitHub's own API, there is no attacker domain to block.
From 19 October one variant added a second job for repositories with self-hosted runners. It searches the host for 31 token formats, private keys, .env files, Terraform state and shell history, reads every process's environment, uses the Docker socket to search the host's root filesystem, and collects cloud instance credentials. Wiz notes that this job is buggy and misses some of what it looks for.
Where it works and where it does not
Since 28 July, GitHub has held potentially malicious workflow runs in public repositories until a collaborator approves them, and every public Gheist run Wiz checked was stuck at that hold. The hold does not apply to private repositories or GitHub Enterprise Server. There, audit logs showed runs on ci-sync branches completing and being deleted by the "gheist" user agent, and in enterprise organisations Wiz saw cloud credentials from the same organisation used hours later to enumerate AWS Secrets Manager.
The same compromised accounts appear in other activity: lure repositories distributing malware, attempts to use Actions for free compute, and victims also infected with PolinRider. Wiz traces the tokens to infostealer logs and to earlier supply-chain compromises.
What security teams should do
Wiz's advice:
- look for ci-sync- branches whose only change adds .github/workflows/sync-check.yml, and in audit logs for runs deleted within a minute of finishing or any activity from the gheist user agent;
- if workflows use OIDC to reach the cloud, check cloud audit logs for role assumptions whose subject includes a ci-sync- branch;
- on self-hosted runners, look for the files .gdmp_pat, .gdmp_patg and .gdmp_sweep;
- rotate every repository, organisation and environment secret available to an affected repository, since deleting the run does not undo the exposure, and revoke the tokens that pushed the commits;
- do not approve held sync-check runs.
To harden, Wiz recommends required reviewers on environments with production secrets, OIDC trust policies scoped to protected branches, GitHub Apps or short-lived fine-grained tokens instead of personal access tokens, and ephemeral self-hosted runners without Docker socket access.
Related on TrustList:
- Wiz ties break-ins at 14 AI developer platforms to stolen model keys
- PingCAP breach: TiDB Cloud users told to replace API keys now
- Alpine rebuilds GitLab after CVE-2026-97229 attempt; the fix needs patched Ruby
Sources
Categories & features
- DevOps
- Credential Management
- Application Security
- Cloud Security
- Threat Intelligence
- Cybersecurity
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
The GitLab fix for the CVSS 10.0 regex flaw is a patch to Ruby that GitLab applies only in its own packages, so installations running their own Ruby also need Ruby 3.3.13, 3.4.12 or 4.0.9.
The Jenkins project's 28 October advisory fixes 13 high and 2 medium flaws in core, among them deserialization code execution, an unauthenticated admin-account race in the setup wizard and agent launchers that skip admin approval.
A hijacked contributor account replaced the macOS, Linux and Windows binaries of five releases on GitHub; v0.16.0 is a clean, attested rebuild, and anyone who installed in the window should rotate secrets.
We classified our 198 security advisories from 22 September to 28 October. 64 described attacks already under way, edge devices led, half of KEV listings came within five days of a fix, and the CVSS score was a weak guide to which ones mattered.
Four AWS bulletins on 22 October: Strands Agents A2A servers leaked conversation history between callers, SAM CLI's Terraform hook allowed command injection, Athena exposed query text, and aws-c-mqtt could crash clients.
Attackers used stolen GitHub tokens from three accounts to publish poisoned builds for about two hours. Pin 1.106.2, clean the files by hand before running Python, and rotate every key the proxy held.
A new GhostAction wave used hijacked maintainer accounts, including the author of an Uber repository, to plant a credential-stealing workflow. Socket counts more than 500 accounts since 7 October.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More DevOpsThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.