Remote Code Execution via Model Deserialization in InvokeAI
InvokeAI patched a remote code execution vulnerability (CVE-2024-12029) in its model deserialization process that could allow execution of arbitrary code through malicious pickle files. The fix improves validation by failing scans on detection errors and enabling scanning by default.
What happened
InvokeAI contained a remote code execution vulnerability in its model deserialization process. The application failed to properly validate pickle files and other model formats during loading, which could allow execution of arbitrary code.
A fix was implemented to fail scans when InvalidMagicError is detected by picklescan, to treat scan errors as failures when checking for infected files, and to enable scanning by default when reading checkpoint metadata.
Governance controls
The vulnerability was addressed through code changes to the model loading and import processes, ensuring that file scans properly reject models with detection errors and enabling scanning by default for model metadata reads.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.