Ml-Logger (<= Commit Acf255bade5be6ad88d90735c8367b28cbe3a743) - CraftedSignal Threat Feedhttps://feed.craftedsignal.io/products/ml-logger--commit-acf255bade5be6ad88d90735c8367b28cbe3a743/Trending threats, MITRE ATT&CK coverage, and detection metadata. Fed continuously.Hugoenhello@craftedsignal.iohello@craftedsignal.ioFri, 16 Oct 2026 19:44:16 +0000https://feed.craftedsignal.io/favicon-32x32.pngCraftedSignal Threat Feedhttps://feed.craftedsignal.io/3232https://feed.craftedsignal.io/favicon.svgUnauthenticated Arbitrary File Read in ml-logger via CVE-2025-10952https://feed.craftedsignal.io/briefs/2026-10-ml-logger-afr/Fri, 16 Oct 2026 19:44:16 +0000hello@craftedsignal.iohttps://feed.craftedsignal.io/briefs/2026-10-ml-logger-afr/An unauthenticated arbitrary file read vulnerability (CVE-2025-10952) in the ml-logger stream_handler allows remote attackers to enumerate and exfiltrate sensitive files from the host filesystem.CVE-2025-10952 is a critical vulnerability residing in the ml-logger Python library, specifically within the stream_handler component of ml_logger/server.py. The flaw stems from improper input validation where the key parameter, provided via JSON request body, is passed directly to file read operations without adequate sanitization or path restriction checks. An unauthenticated attacker can exploit this by sending specially crafted paths, such as those utilizing double leading slashes (e.g., //root/.ssh/id_rsa), to bypass intended directory constraints. This enables the attacker to read arbitrary files accessible by the system user running the ml-logger process. A functional proof-of-concept exploit is publicly available, enabling unauthenticated file enumeration via the /glob endpoint followed by arbitrary file exfiltration through the /stream endpoint.

Attack Chain

  1. Attacker identifies a target exposing the ml-logger API, typically on port 8081.
  2. Attacker sends a POST request to the /glob endpoint to enumerate the filesystem and identify sensitive files.
  3. Attacker parses the JSON response from the /glob endpoint to locate high-value targets like SSH keys, configuration files, or credentials.
  4. Attacker constructs a malicious request to the /stream endpoint, embedding the targeted file path into the key JSON field.
  5. Attacker prepends double slashes (e.g., //etc/shadow) to the file path to bypass the application's path normalization logic.
  6. The ml-logger service retrieves the requested file from the disk and transmits the content to the attacker in the response body.
  7. Attacker saves the exfiltrated content, such as a private SSH key, to their local machine.
  8. Attacker uses the stolen credentials to establish persistent access or move laterally within the victim environment.

Impact

Successful exploitation allows unauthenticated remote attackers to read sensitive files from the underlying host filesystem. This poses a severe risk of credential theft, including the acquisition of SSH private keys, environment variables, and configuration files containing secrets. If the ml-logger service is running with elevated privileges, the impact extends to full system compromise. The vulnerability affects all versions of geyang/ml-logger up to commit acf255bade5be6ad88d90735c8367b28cbe3a743.

Recommendation

  • Implement an authentication-enabled reverse proxy or API gateway in front of all ml-logger instances to restrict access to trusted sources.
  • Deploy the Sigma rule below to detect path traversal attempts targeting the ml-logger endpoints.
  • Restrict the operating system user account running ml-logger to the minimum necessary permissions to prevent access to sensitive files like ~/.ssh/ or /etc/shadow.
  • Upgrade to a patched version of ml-logger as soon as the vendor provides a formal release addressing the vulnerability.
]]>highadvisoryarbitrary-file-readpath-traversalwebserver