Attack Chain

1. A user calls DiffusionPipeline.from_pretrained with a malicious custom_pipeline pointing to an attacker’s repository (repoB) while setting trust_remote_code=False.
2. The DiffusionPipeline.download() function is invoked, but the trust check is performed against the primary repository (repoA) instead of the attacker’s repository (repoB).
3. Alternatively, the user calls DiffusionPipeline.from_pretrained with a local snapshot directory and a malicious custom_pipeline pointing to an attacker’s repository. The local-path branch bypasses the download() function, thus skipping the trust_remote_code gate.
4. As another alternative, the user calls DiffusionPipeline.from_pretrained with a local snapshot directory containing custom component files (e.g., unet/my_unet_model.py) referenced from model_index.json. The local path bypasses download().
5. The attacker’s pipeline.py or custom component files are loaded as dynamic modules.
6. The attacker’s code is executed, granting the attacker arbitrary code execution privileges on the victim’s machine.
7. The attacker can then perform various malicious activities, such as installing malware, stealing data, or compromising the system.

Impact

Successful exploitation of this vulnerability allows for arbitrary remote code execution on the victim’s machine. This could lead to complete system compromise, data theft, or other malicious activities. All users of diffusers versions before 0.38.0 who call DiffusionPipeline.from_pretrained with custom pipelines are potentially affected.

Recommendation

• Upgrade to diffusers version 0.38.0 or later to remediate the vulnerability. The fix moves the trust_remote_code gate to get_cached_module_file in src/diffusers/utils/dynamic_modules_utils.py (https://github.com/huggingface/diffusers/pull/13448).
• If upgrading is not immediately possible, only call from_pretrained with pretrained_model_name_or_path, custom_pipeline, and local snapshot directories from fully trusted sources that have been audited.
• If a local snapshot is used, inspect it for unexpected *.py files, especially under component subdirectories (unet/, scheduler/, etc.) and at the snapshot root before calling from_pretrained.
• Deploy the Sigma rule to detect execution of unexpected python files.