Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

WeaknessASC-2026-0063Other weakness

Poisoned public datasets

MediumWorkarounds only

Datasets on public hubs can be edited after you first used them or seeded with poisoned samples, and web-scraped data can be bought or planted by attackers.

How to detect it

  • Datasets referenced without a fixed revision
  • Training data scraped from domains anyone can buy or edit

Workarounds

  • Pin dataset revisions and keep hashes
  • Filter and sample-check data before training

Small fractions of poisoned data can implant behavior; provenance matters more than volume.

How to investigate

  1. Compare the dataset revision used with the current one
  2. Re-train or re-evaluate without suspicious samples

Sources

Open it in the Learning Center's vulnerability database