Security context

What an agent needs to avoid regressing past fixes and find the next vuln in this repo.

ludwig-ai/ludwig

              main @ b412419
            

Fixes

CVEs

CRITICAL

Peak severity

20.0%

Coverage

Highlights

Arbitrary Code Execution: 2 prior fixes. Scrutinize any change in this area.

ludwig/schema/llms/base_model.py: most-fixed (2 issues). Treat as high-risk during review.

5 high-severity fixes in this history; regressions here are high-impact.

Recurring patterns

The bug types that recur here, drawn from past fixes, not open vulnerabilities.

Insecure Deserialization: Unsafe loading of serialized objects via automated pickle file dispatch or unconstrained PyTorch deserialization leads directly to remote code execution (CVE-2026-31237, CVE-2026-31238). Ensure absolute deprecation of pickle auto-dispatch and enforce weights_only=True during PyTorch load calls.

Arbitrary Code Execution: Remote code validation flags (like trust_remote_code=True) enabled by default allow external, unvetted model configurations to execute arbitrary payloads during HuggingFace lookup. All remote configuration lookups must explicitly default this parameter to False.

Arbitrary Code Execution: Instantiating causal language models from pre-trained repositories with remote code trust enabled allows remote attackers to run arbitrary code on the host machine. The default behavior must restrict remote model code loading.