clip-finetune-recipes
Warn
Audited by Snyk on Jun 13, 2026
Risk Level: MEDIUM
Full Analysis
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (high risk: 0.75). The required runtime workflow for training uses a streaming WebDataset loader over user-supplied shard URLs/patterns (e.g.,
create_webdataset_loader(shard_pattern="s3://.../{00000..01999}.tar")), which ingests outsider-authored free text from thetxtfields inside those tar shards into the LLM context via the training/evaluation code paths that consume captions as strings.
Issues (1)
W011
MEDIUMThird-party content exposure detected (indirect prompt injection risk).
Audit Metadata