CyberCQRThought leadershipBlogWorking togetherBook a conversation
By Neil Brady, CISSPPublished 15 November 2026

Blog · Software supply chain

The keys nobody changed: 4,837 hijacked GitHub accounts

An attacker used 4,837 hijacked GitHub accounts to steal secrets from software builds. Why it matters to UK firms that never write code, and what to ask.

15 November 2026 · Reading time about 5 minutes

On 10 November an attacker began using 4,837 hijacked GitHub accounts to test which code projects they could change [1]. GitHub is where many organisations store and build their software. Within a day, those accounts had probed 61,764 repositories, and a credential-stealing script ran in at least 1,806 of them [1][2].

The attack didn't break GitHub. It logged in.

Where an account could make changes, the attacker slipped a 1,500-line script into the project's automated build [1]. It collected whatever that build could reach: cloud keys, publishing tokens and keys for artificial intelligence (AI) services [1]. Then it removed its own commit and deleted the logs. Each visit took about a minute [2].

These weren't fresh thefts. In October, a campaign called GhostAction used about 1,900 hijacked accounts in much the same way [3]. Aikido Security found that accounts hit in October were about 1,000 times more likely than other active accounts to turn up again this week [4]. The keys were taken weeks earlier and still opened the door, because nobody revoked them.

I've seen how long a credential can outlive its purpose. A large bank I worked with used certificates from its own internal system to decide which devices could join the network and the Wi-Fi. It had never revoked one. The check asked only whether the bank had issued the certificate, not which device it belonged to. During a pilot, one certificate got more than 1,000 devices on. When I showed that a copy taken from one machine let another straight in, the first response was disbelief. The second was that it was too big to fix without the outsourced network company.

It's the pattern I wrote about with patching. Not changing a credential is a decision too. You just don't get to choose when it costs you.

Most UK SMEs don't run GitHub. Their suppliers do: the agency that built the website, the developer who maintains the booking system, the software firm behind the line-of-business app. Those build systems often hold keys to your cloud, your domain or your customer data. When did you last ask who looks after them?

How the wfx-loot GitHub attack worked

Researchers at Aikido Security, who named the campaign wfx-loot, saw each account open a harmless-looking pull request titled "docs: dependency audit notes" [1]. Where that succeeded, the script went into the automated build process, known as continuous integration (CI). The next run triggered it.

Affected accounts include some linked to Fortune 100 firms, the US General Services Administration and the French government [1]. When BleepingComputer reported the story, 92% of the probe requests were still open [1]. Most owners hadn't noticed.

GitHub still hadn't commented by 14 November [5]. Endor Labs reports that the accounts belong to real developers caught in earlier supply chain attacks [2]. One victim traced it to a leaked token for GitHub's command-line tool [2]. Fly.io, a hosting firm, found that some came from its own GitHub app, exposed in configuration files anyone could read [5].

Why hijacked developer accounts matter if you don't write code

This is a supply chain risk, because a supplier's weakness becomes yours. There's an identity risk too, because the attacker used valid logins rather than a software flaw. And it's an intellectual property risk, because access that can plant a script can also copy your source code.

Build systems and bot accounts are non-human identities: logins used by software rather than people. They rarely get the scrutiny that employee accounts get. In the WebdriverIO project, the attacker pushed code through a bot account allowed to bypass branch protection [2]. Few people review a bot's sign-ins, so no one spots when someone else is using it.

If you cannot detect it, you cannot protect it.

Secrets management: treat keys as assets, not settings

Every token, key and certificate is an identity. Each one needs an owner, an expiry date and a record of where it sits. Large organisations manage them centrally, in tools such as CyberArk, HashiCorp Vault and Venafi. These issue short-lived credentials on demand, rotate them automatically and log every use.

Most SMEs don't need an enterprise platform. Microsoft Azure, Amazon Web Services and Google Cloud each offer their own vault service. The usual gap is that nobody has switched it on, so keys still live in code, spreadsheets and developers' laptops.

What to ask your software suppliers this week

Here's my view, and some will disagree. For most SMEs the risk here isn't their own code. It's credentials they handed to a supplier and never asked about again.

Ask your IT provider and software suppliers:

  • Did wfx-loot touch you? Researchers have published what to look for, including a branch called "wfx-docs" and gaps in build run numbers [1]. A probe request alone means the account was compromised, even if the script never ran [2].
  • Do your build keys expire? A credential issued for a single run and gone an hour later is far less use to a thief than one that lasts for years.
  • Who can change your build files? Edits to these workflows should need review by a named owner. Nobody, human or bot, should push straight to the main branch [2].
  • Is your multi-factor authentication phishing-resistant? Hardware keys or passkeys for anyone who can change code [2].
  • Where do you keep the keys you hold for us, and when were they last rotated? If the answer is "in the code" or "never", start there.

If a supplier was hit, assume every key their builds could reach is stolen. Rotating them isn't enough on its own. The stealer tried to leave its own way back in, such as new trust rules in Microsoft Entra ID named wfx-ci [6]. So rotate everything, then check your cloud accounts for access keys or trust rules created since 10 November [2].

Read more

Sources

  1. [1]BleepingComputer, Hackers compromise 4,800 GitHub accounts to steal cloud credentials, 12 November 2026. https://www.bleepingcomputer.com/news/security/hackers-compromise-4-800-github-accounts-to-steal-cloud-credentials/
  2. [2]Endor Labs, Hijacked GitHub accounts push "wfx" CI secret stealer to thousands of repositories, updated 12 November 2026. https://www.endorlabs.com/learn/github-actions-supply-chain-attack-targeting-ci-secrets-at-scale
  3. [3]The New Stack, Hackers hijacked 1,900 GitHub accounts. Then they went after 30,000 repos, 15 October 2026. https://thenewstack.io/hackers-hijacked-1900-github-accounts-then-they-went-after-30000-repos/
  4. [4]The Hacker News, Attacker used 4,837 hijacked GitHub accounts to run a CI secret stealer in 1,806+ repositories, 12 November 2026. https://thehackernews.com/2026/11/4837-hijacked-github-accounts-ran-ci.html
  5. [5]heise online, Stolen tokens: Secret stealer targets thousands of GitHub repositories, 14 November 2026. https://www.heise.de/en/news/Stolen-tokens-Secret-stealer-targets-thousands-of-GitHub-repositories-11524944.html
  6. [6]Cloud Security Alliance, The wfx campaign: hijacked GitHub accounts harvest CI secrets, 13 November 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-wfx-github-ci-secret-stealer-20261112-csa

All blog posts