Tensorlake reports malicious GitHub workflow runs after an upstream breach

Tensorlake says an internal automation app running on an upstream provider started sandboxes under engineering team accounts. Those sandboxes had access to Tensorlake GitHub repositories. An attacker pushed malicious code to branches and triggered GitHub Actions to probe for build and deployment secrets. Tensorlake says it found no evidence of access to critical systems beyond the probing and reports that customer API keys and sandboxes were unaffected. The company has not published affected branch names, workflow identifiers or evidence that particular secrets were retrieved.

The founder’s initial public post named Railway as the app host. In the edited statement, Choudhury referred to an upstream provider and said a prior breach there had exposed environment variables that Tensorlake had left unrotated for the internal app. The follow-up also attributes the exposure to the upstream breach. The initial post and follow-up are available in the disclosure’s public copy and the reply’s public copy. A public copy of the edited statement preserves the later wording. The relationship to Railway’s October incident has not been independently verified. The earlier Tensorlake npm advisory covers a different event.

Tensorlake says it rotated its secrets and removed the app from the upstream host. It emailed paying customers and asked customers to rotate Tensorlake API keys as a precaution. Further details are expected by email and in the Tensorlake trust center.