Assessor staff find a malicious prompt template circulating on the internal wiki for valuation assistants. How should the assessment treat it?
Select an answer to reveal the explanation.
Short Explanation
A bad prompt on the wiki is supply chain too—templates travel like shared macros. Review and lock them down.
Full Explanation
A malicious prompt template circulating on the assessor internal wiki for valuation assistants should be treated as a supply-chain artifact: shared prompts are configuration that alters model behavior at scale across civic assistants. Unreviewed or attacker-edited templates can bias valuations, exfiltrate parcel data via instructions, or weaken safety policies without changing application code. Assessments recommend curated or signed prompt libraries, change control, access restrictions on wiki pages that feed production assistants, and integrity checks before templates are loaded. Office décor does not secure prompt supply.
Treating shared prompt templates as documentation aids requiring readability checks and formatting standards, treating them as model output requiring accuracy sampling and valuation tolerance checks, and storing them in assessment workpapers with access logging and retention scheduling all describe how templates might be handled, but none of them states the assessment's required conclusion — that templates themselves must be treated as supply-chain artifacts needing integrity and review controls.
Exam caveat: prompt templates may be copied into multiple bots; fixing one wiki page is insufficient if forks exist in tickets, chats, or local files—inventory all load paths. Operational check: move production valuation prompts into a version-controlled, reviewed library with restricted editors; hash-check templates at assistant startup; remove the malicious wiki copy; and sample-run valuation scenarios to confirm behavior matches the approved prompt revision.