CVE-2026-1839 | Arbitrary Code Execution via Unsafe torch.load() in Trainer Checkpoint Loading in huggingface/transformers

A vulnerability in the HuggingFace Transformers library, specifically in the `Trainer` class, allows for arbitrary code execution. The `_load_rng_state()` method in `src/transformers/trainer.py` at line 3059 calls `torch.load()` without the `weights_only=True` parameter. This issue affects all versions of the library supporting `torch>=2.2` when used with PyTorch versions below 2.6, as the `safe_globals()` context manager provides no protection in these versions. An attacker can exploit this vulnerability by supplying a malicious checkpoint file, such as `rng_state.pth`, which can execute arbitrary code when loaded. The issue is resolved in version v5.0.0rc3.

Published: 2026-04-07 Last update: 2026-04-07 Assigner: [email protected] Source: [email protected]
NVD Status：Undergoing Analysis，CVE State: published
View at NVD, CVE.org, EUVD

CVE-2026-1839 is rated Low Risk (27.8/100): CVSS Medium severity, with low exploitation likelihood (EPSS 0.02%). Monitor for updates and reassess as exploit intelligence or EPSS changes.

Risk is dynamic; we continuously reassess and refresh what is shown on this page as upstream context changes.

Exploit prediction scoring system (EPSS) score for CVE-2026-1839

EPSS lead: Daily EPSS estimates relative likelihood of exploitation; percentile ranks this CVE among scored vulnerabilities (higher = more severe relative rank).

# Date Old EPSS score New EPSS score Delta (New - Old)
1 2026-04-07 — 0.02% —

Full EPSS history (1 record total)

Common vulnerability scoring system (CVSS) metrics for CVE-2026-1839

CVSS metrics for this CVE.

Base score Version Severity Vector Exploitability Impact Score source
6.5 3.0 MEDIUM
CVSS:3.0/AV:L/AC:H/PR:N/UI:R/S:U/C:H/I:L/A:H Click to expand
Attack vector (AV:L)
They already need access on the box, or another person has to do something wrong; it’s not a remote drive-by.
Attack complexity (AC:H)
Even with access, the exploit needs extra luck, timing, or a fussy environment to actually work.
Privileges required (PR:N)
No account or special rights needed—anonymous or random user is enough.
User interaction (UI:R)
A real person has to do something—click, install, enable—otherwise it doesn’t land.
Scope (S:U)
Damage stays in the same “trust bubble” as the broken component—no big spill into unrelated systems.
Confidentiality (C:H)
Serious risk that confidential data gets exposed in a big way.
Integrity (I:L)
Attackers could change some data, but it’s limited—not everything goes.
Availability (A:H)
Could take the service down hard or make it unusable for people who depend on it.
 1.0 5.5 [email protected]

Weakness enumeration for CVE-2026-1839

GitHub Security Advisory for CVE-2026-1839

GHSA-69w3-r845-3855 · Severity: medium · Ecosystem: pip — HuggingFace Transformers allows for arbitrary code execution in the `Trainer` class

OS Trackers for CVE-2026-1839

vendor priority summary link
redhat medium — https://access.redhat.com/security/cve/CVE-2026-1839

Affected software / configurations for CVE-2026-1839

Vendor Product Version Raw CPE
No affected products in dataset.

References for CVE-2026-1839

cvelogic Threat Intelligence