Sample question
A company periodically fine-tunes its code assistant on internal repositories. A departing developer commits dozens of files containing a hardcoded login bypass, and after the next fine-tuning run the assistant starts suggesting that bypass to colleagues. Which of the following threats does this describe?
Pick an answer to see why each option is right or wrong.
Takeaway: tampered training data is poisoning; tampered input at use time is prompt injection. Sources: NIST AI 100-2 E2025 · OWASP Top 10 for LLM Applications: LLM04 Data and Model Poisoning