Hugging Face Access Token

Detects Hugging Face user access tokens (hf_ prefix), used to authenticate to the Hugging Face Hub API. A leaked token grants access to private models, datasets and Spaces, and can incur inference billing.

Type

regex

Engine

universal

Confidence

high

Confidence justification

High confidence: the distinctive hf_ prefix with a fixed 34-character body is highly specific to Hugging Face access tokens.

Jurisdictions

global

Regulations

Criminal Code Act 1995 (Cth)

Frameworks

CIS Controls, ISO 27001, NIST CSF, SOC 2

Data categories

credentials, security

Scope

narrow

Risk rating

8

Platform compatibility

Purview: Compatible, GCP DLP: Unsupported, Macie: Unsupported, Zscaler: Compatible, Palo Alto: Unsupported, Netskope: Unsupported

Pattern

(?<![A-Za-z0-9_])hf_[A-Za-z0-9]{34}(?![A-Za-z0-9])

Corroborative evidence keywords

hugging face, huggingface, HF_TOKEN, access token, api token, transformers, inference api

Proximity: 300 characters

Should match

• hf_ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefgh — Hugging Face access token, hf_ prefix + 34 chars
• HF_TOKEN=hf_zyxwvutsrqponmlkjihgfedcba01234567 — Labelled Hugging Face token in an env assignment
• hf_aB3dE6gH9jK2mN5pQ8sT1uV4wX7yZ0pLkJ — Mixed-case Hugging Face token

Should not match

• hf_ABCDEFGHIJKLMNOP — Too short to be a valid Hugging Face token
• hg_ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefgh — Wrong prefix (hg_ instead of hf_)
• load the model with your hugging face access token set as HF_TOKEN — Prose mention without a token value

Known false positives

• Documentation or examples showing placeholder hf_ strings. Mitigation: Require corroborative keywords and check for placeholder markers (example, xxxx).

References

• User access tokens - Hugging Face Hub