VDB
KO

MAL-2026-10919

Malicious code in ml-core-airflow-auth (PyPI)

Details

--- _-= Per source details. Do not edit below this line.=-_

## Source: amazon-inspector (0a543a55b463967fa7f6b3c1462ecb38735072a4f4f47bf5f9b732e089a64b8a) The package advertises itself as an 'auth token handler' but its public API (src/ml_core_airflow_auth/__init__.py) exposes only a trivial in-memory TokenStore stub unrelated to the ~3300 LOC of hidden telemetry code shipped alongside. The setup.py custom install command copies telemetry.pth into site-packages; that.pth file contains `import _telemetry_init`, causing every subsequent Python interpreter start on the host to auto-load _telemetry_init._bootstrap, which spawns a daemon thread that constructs an analytics Client (Envelope, HttpTransport, ServiceDiscovery, DataScrubber, SessionTracker) and calls client.track('session_start') with the machine's hostname, OS, Python version, and cpu_count. The reporting endpoint is not a static URL; _telemetry_transport.py's ServiceDiscovery opens a raw UDP DNS socket with hardcoded fallback resolvers 8.8.8.8 and 1.1.1.1, queries TXT records under `0.<domain>`... `N.<domain>`, and reconstructs the destination via `base64.b64decode(''.join(segments))`, letting the operator change the exfiltration destination post-install without republishing. The name, generic 'Platform Engineering' author, and mismatch between the declared purpose and the hidden persistent telemetry framework are consistent with dependency-confusion targeting corporate ML/Airflow environments.

## Source: kam193 (1b104d8446715fc7f59f0274ca3e655e6818b45edd837c24f78e857c917b2e6f) Package presents little functionality, but excessive fake 'telemetry' module. This fake telemetry is used to download and run malicious executables. Code is designed to survive different blocks: first, there is an attempt to download the executable from one of five Cloudflare Workers. If it's not successful, the code falls back to download using DNS: first, it gets a TXT record from one of c.*.dl.well1[.]site domains, depending on the system. This record returns a number, which is then used to iterate over domains in the form <0...n>.*.dl.well1[.]site and reconstruct the encoded executable from their TXT records. The downloaded binary is then executed and removed afterward. Using a PTH file ensures persistence and runs on every Python start. In this campaign, versions 0.0.1 hold disarmed code (without the necessary configuration), which is completed in further updates.

This is a continuation of the 2026-07-haproxy-config-client campaign.

---

Category: MALICIOUS - The campaign has clearly malicious intent, like infostealers.

Campaign: 2026-07-andreiiiiiii_i

Reasons (based on the campaign):

- The package contains code to exfiltrate basic data from the system, like IP or username. It has a limited risk.

- The package overrides the install command in setup.py to execute malicious code during installation.

- Downloads and executes a remote executable.

- covering-tracks

- persistence

- abuses-pth

- data-stored-in-dns

Are you affected?

Enter the version of the package you're using.

Affected packages

PyPI / ml-core-airflow-auth

No fixed version published yet for ml-core-airflow-auth (pip). Pin to a known-safe version or switch to an alternative.

References