Hugging Face user access token

ID

huggingface_token

Severity

high

Remediation Complexity

trivial

Remediation Risk

high

Remediation Effort

low

Vendor

Hugging Face

Family

API Token

Description

Hugging Face hosts models, datasets and Spaces. A user access token authenticates the Hub API, git operations over HTTPS, and the Inference API.

hf_ followed by 34 alphanumeric characters.

Security

A leaked Hugging Face key grants an attacker the same access as the application that owns it:

  • Unauthorized usage is billed to your account, and automated high-volume calls can run up a large charge before anyone notices.

  • Whatever is sent through the leaked key — prompts, documents, query results — is exposed to whoever holds it.

  • A write-scoped token can push to your models, datasets and Spaces. That turns the leak into a supply-chain risk for everyone who pulls those artifacts.

  • The key does not expire on its own, so the exposure lasts until it is revoked.

Examples

The following is a leak of a Hugging Face key:

HF_TOKEN=hf_NSnkp0Tu...DXlo5L

Mitigation / Fix

  1. Follow your policy for handling leaked secrets, which normally requires revoking the key. Go to [Hugging Face > Settings > Access Tokens], find the leaked token and use "Invalidate and refresh", or delete it and create a replacement with the narrowest scope that still works. Prefer fine-grained tokens over unscoped read/write ones.

    Revoke the leaked key, do not merely disable it.

  2. Remove the leaked key from the source code or committed configuration file, and replace its usages with the new value. Environment variables, local files or secret vaults could be used for passing the key, instead of hardcoding the value, as documented in How to Prevent Hard-Coded Secrets.

  3. Review the usage and audit logs for the key, to confirm that it was not used by unintended actors while it was exposed.

  4. Apply the usual precautions for API keys: never commit them, pass them through environment variables or a secret manager, use a separate key per environment, monitor usage for unusual spikes, and rotate them periodically.

You should consider any sensitive data in commits with secrets as compromised.

Remember that secrets may be removed from history in your projects, but not in other users' cloned or forked repositories.

Reference