CVE-2026-44827
HighCVSS 8.8Exploitation Probability (EPSS)
Elevated risk54th percentile - higher than 54% of all known CVEs
Summary
In diffusers prior to 0.38.0, the `_resolve_custom_pipeline_and_cls` function defaults `custom_pipeline` to `None.py`, allowing a malicious file from a Hugging Face Hub repository to be downloaded and executed during a standard `from_pretrained()` call. This bypasses the `trust_remote_code=True` safeguard, enabling silent remote code execution.
Risk Assessment
An attacker could achieve arbitrary code execution in the victim's environment, leading to full system compromise, data theft, or further escalation.
Recommendation
Upgrade diffusers to version 0.38.0 or later. Avoid loading models from untrusted Hugging Face Hub repositories.
Other vulnerabilities in diffusers
See all- CVE-2026-65920Medium
A path traversal vulnerability was found in Diffusers up to version 0.39.0 in the _get_checkpoint_shard_files function. Attackers can supply malicious weight_map values in model index JSON, containing ../ sequences or absolute paths, allowing reading of safetensors files outside the intended directory.
- CVE-2026-45804High
A vulnerability in the Diffusers library before version 0.38.0 allows bypassing the trust_remote_code guard. The download() method validates model_index.json and custom pipeline code but later loads from a cached folder that can change, enabling execution of malicious .py code from a Hub repository without requiring trust_remote_code=True.
- CVE-2026-44513High
The Diffusers library before version 0.38.0 contains a vulnerability that bypasses the trust_remote_code protection. An attacker can execute arbitrary remote code even when trust_remote_code is set to False or omitted. The issue stems from a flawed implementation of the security check in the download() function, which can be bypassed through alternative code paths.
Original NVD description (English source)
Diffusers is the a library for pretrained diffusion models. Prior to 0.38.0, diffusers 0.37.0 allows remote code execution without the trust_remote_code=True safeguard when loading pipelines from Hugging Face Hub repositories. The _resolve_custom_pipeline_and_cls function in pipeline_loading_utils.py performs string interpolation on the custom_pipeline parameter using f"{custom_pipeline}.py". When custom_pipeline is not supplied by the user, it defaults to None, which Python interpolates as the literal string "None.py". If an attacker publishes a Hub repository containing a file named None.py with a class that subclasses DiffusionPipeline, the file is automatically downloaded and executed during a standard DiffusionPipeline.from_pretrained() call with no additional keyword arguments. The trust_remote_code check in DiffusionPipeline.download() is bypassed because it evaluates custom_pipeline is not None as False (since the kwarg was never supplied), while the downstream code path that actually loads the module resolves the None value into a valid filename. An attacker can achieve silent arbitrary code execution by publishing a malicious model repository with a None.py file and a standard-looking model_index.json that references a legitimate pipeline class name, requiring only that a victim calls from_pretrained on the repository. This vulnerability is fixed in 0.38.0.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS