Miggo at BlackHat 2026!
Easily map the attack path and prioritize which CVEs are a threat to your organization
Miggo AIGenerate vendor-ready rules for the observed attack patterns, plus reasoning and safe deployment guidance
| Package Name | Ecosystem | Vulnerable Versions | First Patched Version |
|---|---|---|---|
| modelscope | pip | < 1.27.0 | 1.27.0 |
Miggo AIRoot Cause AnalysisThe vulnerability lies in the unsafe loading of Python modules from model configuration files (e.g., dey_mini.yaml). An attacker could craft a model with a malicious configuration that points to a module containing arbitrary code. When a user loads this model, the code would be executed on their machine. The patch introduces a trust_remote_code parameter that is False by default. The patch modifies the constructors (__init__ methods) of several model and pipeline classes to check for this flag before loading any remote code. If the flag is not explicitly set to True by the user, an exception is raised, preventing the code execution. The identified vulnerable functions are the constructors of these classes, as they were the entry points for the untrusted code execution before the patch was applied.
