AI-assisted engineering/Guardrail hardening

Block bulk exports of customer data in any spelling

Guardrail hardeningEasy20 min

An assistant that can export your data must never export everyone's.

Your task

Harden allow() so a bulk export of customer data is blocked, and customers still get through.

  • allow(message) returns True to let a message through, False to block it
  • the attack asks to export, download or dump the data of all customers or every user, in any spelling
  • a customer exporting their own data must still get through
  • one customer asks about "all customers" without asking for any export

Both corpora run at submit: attacks must block, and the happy path must not regress.