# Tensorlake’s Malicious npm Release Targeted Developer Credentials

Published 2026-10-09 · AI.info Team · AI.info

> A compromised maintainer account published malicious npm version 0.5.144 of Tensorlake on October 8, 2026, the company says. The install-time payload targeted developer credentials and AI-tool configuration, while Tensorlake says the release was not authorized.

For about 1 hour and 40 minutes on October 8, npm hosted a malicious release of [Tensorlake](https://github.com/tensorlakeai/tensorlake/security/advisories/GHSA-8g63-53c9-65j2)’s TypeScript SDK. Version 0.5.144 appeared between 01:12 and 02:54 UTC after an attacker used a compromised maintainer account to push code to Tensorlake’s repository and publish through its release workflow, the company said. The release even carried valid npm provenance, but that did not make it authorized.

> “This version was not an authorized Tensorlake release.”
>
> [Tensorlake security advisory](https://github.com/tensorlakeai/tensorlake/security/advisories/GHSA-8g63-53c9-65j2)

The npm release is the affected distribution, and Tensorlake’s PyPI and crates.io packages, along with its command-line tool, were not affected, according to the advisory. It lists version 0.5.144 for the main package and platform-specific native packages. No use of that version had been seen through the SDK usage Tensorlake monitors, the company also said.

## Version 0.5.144 ran during installation

The compromised package used a `preinstall` hook to run an obfuscated JavaScript payload, so its code could execute as part of installation rather than waiting for an application to import the SDK. The installer launched the payload through Bun and could download Bun 1.3.13 when it was not already present, according to Sonatype’s analysis. The code’s credential collection had little to do with the SDK’s stated work with sandboxes, repositories and cloud services.

The payload targeted [GitHub](https://ai.info/companies/github) and npm tokens, [AWS](https://ai.info/companies/amazon) and [Google Cloud](https://ai.info/companies/google-cloud) credentials, Kubernetes and Vault tokens, SSH and GPG keys, environment files, browser-saved passwords, cryptocurrency wallets and AI-tool and MCP configuration, according to Tensorlake’s advisory. Sonatype’s review names files associated with Claude, [Cursor](https://ai.info/companies/anysphere-inc), Kiro, Codex, Gemini and Windsurf. The payload also polled a command channel and evaluated JavaScript supplied remotely, according to Sonatype.

## Stolen access could reach other repositories

The malware used stolen tokens to republish itself into other npm packages and repositories the victim could write to, the advisory says. Sonatype found code that could create a public GitHub repository and commit collected data there; one code path could include a stolen GitHub token in the uploaded data and commit message. Its analysis also found logic to add `.claude` and `.vscode` files to accessible repositories, potentially exposing other developers who later use them.

The advisory describes a persistent service called `gh-token-monitor` that deletes the user’s home directory if the stolen GitHub token is revoked. Users should remove that service before revoking tokens, Tensorlake says. The destructive behavior raises the stakes of the response: simply rotating credentials first could trigger the condition the malware was designed to exploit.

## Confirmed reach remains narrower than the payload’s capabilities

Sonatype found that the install hook exits when it detects several specific CI environment markers, including certain GitHub Actions and GitLab CI variables. That behavior does not establish that CI systems were outside the attacker’s aims, and it does not show that every CI environment was spared. Teams should treat a machine or CI system as compromised if the version was installed with scripts enabled, according to Tensorlake’s advisory.

Neither the advisory nor Sonatype’s analysis confirms that collected data was published, repositories were modified or downstream systems were affected. The potential for those outcomes is not evidence that they occurred. Sonatype is continuing its investigation as additional components and impact are confirmed.

## Checking lockfiles and build logs

Tensorlake advises checking project dependencies for `tensorlake@0.5.144` and platform-specific `tensorlake-native-*` packages at the same version, and reviewing build records for installations between 01:12 and 02:54 UTC on October 8. Its advisory identifies version 0.5.143 and earlier as unaffected and recommends moving back to 0.5.143, or to 0.5.145 once published at the time of the advisory.

For systems where the malicious version ran, the company recommends disconnecting the machine, removing the token-monitor service, then rotating accessible credentials from a separate clean machine. It also advises checking npm packages and repositories those credentials could publish to or modify, then rebuilding the affected machine from a trusted image. Removing the package alone cannot revoke credentials it may already have read.

## Sources

1. Original source [Tensorlake via GitHub Security Advisories](https://github.com/tensorlakeai/tensorlake/security/advisories/GHSA-8g63-53c9-65j2) github.com

Other pages read

2. [Sonatype](https://www.sonatype.com/blog/hijacked-tensorlake-npm-turned-install-into-credential-risk) sonatype.com

## Related stories

- [Fake Apex Accelerator Routed AI Coding Traffic to Its Servers](https://ai.info/news/fake-apex-accelerator-routed-ai-coding-traffic)
- [Claude Desktop Flaw Ran Malicious Git Executables in Repositories](https://ai.info/news/claude-desktop-windows-malicious-git-executable-flaw)
- [ARTEX Developer Ends Public Releases After South Korean Bank Attacks](https://ai.info/news/artex-developer-ends-public-releases-south-korean-bank-attacks)
- [State of AI Report Details Agents’ Breach of Hugging Face](https://ai.info/news/state-of-ai-report-agents-hugging-face-breach)
- [Mohiuddin Finds MCP Security Bugs Across Five Organizations](https://ai.info/news/mohiuddin-mcp-security-bugs-five-organizations)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/tensorlake-malicious-npm-release-developer-credentials
