In October, the web learned that eight intermediate certificates issued by eMudhra, an Indian certificate authority, had private keys that were almost all zeros. Only the last byte varied, so anyone could work them out. Browsers blocked them quickly. Build machines have been slower.
What happened with eMudhra?
According to InfoQ, there was no breach in the usual sense. The keys were weak enough to enumerate, and matching public keys were found on live CAs. Six of the eight intermediates were trusted for TLS by Apple, Chrome, Microsoft and Mozilla, and two for code signing by Microsoft. With those keys, an attacker on the network path could issue a valid-looking certificate for any site.
Browsers blocked the intermediates through their revocation lists. Most other software cannot block a single intermediate, so many projects removed the whole emSign ECC Root CA - G3 instead. Fixed CA bundles have shipped in Debian, Ubuntu, Fedora, Arch and Gentoo, and in Go, certifi for pip, rustls-webpki, webpki-roots, Electron and uv.
What is wrong on the macOS runners?
An issue opened on 2 November in GitHub's runner-images repository reports that all public macOS images, on GitHub Actions and Azure DevOps, still trust the affected roots. Two things cause it:
The system bundle at
/etc/ssl/cert.pemstill contains the affected roots. Apple has updated its Network.framework trust but not that file.Homebrew has shipped patched
curl,gitandca-certificates, but the runner images turn Homebrew updates off, so the old versions stay.
A test in the issue shows connections using the compromised chain succeeding on a macOS runner while Windows and Linux runners reject them. One commenter found that Apple's own /usr/bin/python3 accepted the bad chain even with SSL_CERT_FILE set. At the time of writing the issue was open with no maintainer response.
Why should CI teams care?
A build runner downloads dependencies, pulls images, pushes artefacts and talks to cloud APIs with real credentials. If an attacker can sit on that network path with a trusted certificate, they can read secrets in transit or swap a download. The risk on GitHub's hosted network is low, but the whole point of TLS is not having to rely on that.
What should you do?
Repair the trust store on macOS jobs. Commenters confirmed that fixing
/etc/ssl/cert.pemworks. Do it in an early step, before any network calls, until the images are fixed.Update Homebrew tools in sensitive jobs. Upgrade
curl,gitandca-certificatesexplicitly in jobs that handle release credentials.Do not rely on environment variables. Some tools ignore
SSL_CERT_FILE, as the Python test showed.Find vendored CA bundles. Apps and images that ship their own
cacertsfile never see OS updates. Search your images and repos for them and refresh lockfiles that pin old certifi or webpki-roots.Check self-hosted runners and base images too. Rebuild them with current ca-certificates packages.
Expect some breakage. Removing the emSign G3 root breaks legitimate certificates issued under it. If an internal or partner service fails TLS now, that is likely why.
Quick answers
What happened with eMudhra certificates?
Private keys for eight eMudhra intermediate CA certificates were found to be guessable because they were almost all zeros. Six were trusted for TLS, so attackers could potentially issue certificates for any site.
Are GitHub Actions macOS runners affected by the eMudhra compromise?
An open issue reports that public macOS runner images still trust the affected roots through /etc/ssl/cert.pem and outdated Homebrew tools, while Windows and Linux runners reject the compromised chain.
How do I fix eMudhra trust in CI?
Repair or replace /etc/ssl/cert.pem on macOS runners, update curl, git and ca-certificates, rebuild base images with updated CA packages, and find and refresh any vendored CA bundles in your applications.







