AI Safety: The Audit Checklist

Use this checklist before accepting any AI code:

• Spelling: Are column names spelled exactly as they appear in the data?

• Case Sensitivity: Is it Age or age? (R sees these as different things).

• Missingness: Did the AI remember na.rm = TRUE for math?

• Privacy: Did the AI try to print the whole dataset (potentially exposing patient IDs)?

TipPrompt-engineering tip

When asking AI for help, explicitly say: “Do not change the logic or intended result of my code. Explain any suggested change before rewriting it.” This keeps the AI in a reviewing role instead of letting it quietly rewrite your analysis.