Security context

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

embeddings-benchmark/mteb

              main @ 55ceddb
            

Fixes

CVEs

HIGH

Peak severity

Highlights

Code Execution: 1 prior fix. Scrutinize any change in this area.

mteb/_create_dataloaders.py: most-fixed (1 issue). Treat as high-risk during review.

1 high-severity fix in this history; regressions here are high-impact.

Recurring patterns

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

Code Execution: The MIRACL dataset loader bypassed user-configured safety settings and forced trust_remote_code=True, potentially enabling arbitrary remote code execution when loading untrusted datasets. Developers must explicitly propagate the user-defined trust_remote_code setting down to all downstream dataset loaders.

Denial of Service: The video ingestion loader lacked bounds checking when sampling frames, creating an out-of-memory vulnerability when parsing large or malformed video inputs. This was mitigated by introducing a maximum frame cap on FPS-based sampling.