CVE-2026-4372

What it is

A critical remote code execution vulnerability exists in all versions of the HuggingFace transformers library prior to version 5.3.0. The vulnerability allows an attacker to craft a malicious `config.json` file containing the `_attn_implementation_internal` field set to an attacker-controlled HuggingFace Hub repository ID. When a victim loads this model using the standard `AutoModelForCausalLM.from_pretrained()` API, the library downloads and executes arbitrary Python code from the attacker's repository with the victim's full OS privileges. This issue arises due to unfiltered deserialization of configuration attributes, insufficient sanitization of internal fields, and unsandboxed execution of downloaded kernels. The vulnerability bypasses the `trust_remote_code` security mechanism, is invisible to the victim, and exploits the standard documented usage pattern, making it particularly severe. Users are advised to upgrade to version 5.3.0 or later to mitigate this issue.

At a glance

Exploitability

Can be exploited locally on the machine without logging in, but needs a user to act. High risk of data exposure, data tampering, service disruption.

Exploit odds

0.7% chance of exploitation in the next 30 days (EPSS).

Fix

A patch, advisory or fixed version is referenced. Upgrade or apply it.

Timeline

1. 2026-05-24 14:16 UTC PublishedDisclosed publicly on CVE.org and NVD
2. 2026-07-23 17:10 UTC Last updatedMost recent change to the record

Details

Published

2026-05-24 14:16 UTC

Last modified

2026-07-23 17:10 UTC

Exploit probability

0.7% · EPSS, next 30 days

Weakness

CWE-1066

Record status

Analyzed

Affected

• Huggingface Transformers < 5.3.0

Attack profile CVSS 3.0 · 7.8

Attack vector

Local

Complexity

Low

Privileges needed

None

User interaction

Required

Scope

Unchanged

Confidentiality

High

Integrity

High

Availability

High