Governance and Defensive Engineering

COASP · 32 questions

• A city AI test-and-evaluation plan for the permitting copilot includes explicit go/no-go gates before public release. What governance practice is that?
• County compliance links assurance cases to procurement clauses for the 311 LLM vendor. Why does that matter?
• A public-health AI risk committee maps controls to NIST AI RMF-style functions for auditors. What governance skill does that demonstrate?
• A transit authority needs a purple-team AI security report that executives can act on. What should the report emphasize?
• An elections office applies heightened test-and-evaluation before any generative answers on voting topics. What principle guides that choice?
• A water utility compliance matrix assigns who signs off on AI safety evidence before release. What is being established?
• A housing authority AI policy references ISO/IEC 42001-style management-system practices. What is the intent?
• A clerk office wants to analyze how the records-assistant LLM could fail in ways that harm residents or records integrity. Which governance approach best fits?
• City council will only approve launching the permitting copilot after purple-team issues are fixed. What governance demand does that embody?
• A resident-data leak scare involving the 311 assistant requires one shared severity model. What should SOC and the privacy office do?
• Before the permitting AI goes live, the CISO wants one gate covering poison tests, prompt suites, and plugin review. What should the team build?
• A county attorney reviews plans for an authorized AI red team against a live clerk assistant. What governance issue must be settled first?
• After hardening the open-data RAG portal, residual risk remains. What should the project present to the risk committee?
• Emergency management runs AI tabletops for the crisis-comms assistant but never documents lessons. What governance practice is missing?
• A library ethics board will not approve the chatbot rollout until bias and representation tests are planned. What governance principle does that enforce?
• Finance wants the tax-FAQ chatbot’s controls mapped to specific statutory and policy obligations. What governance artifact is needed?
• The 311 bot needs defenses against instruction-override style prompts without banning every unusual phrasing residents use. What defensive engineering choice fits?
• A permitting agent can call tools that touch case systems. What mitigation best limits blast radius if the agent is steered poorly?
• Public-health analysts notice triage-assistant recommendations shifting week over week without an announced model change. What defensive control should they emphasize?
• The clerk bot shows signs of a suspected prompt-injection incident during business hours. What should the blue-team playbook prioritize?
• Housing documents in a RAG corpus include sensitive case notes. What hardening package best fits authorized defensive engineering?
• Benefits-eligibility explanations will go public soon. What defensive engineering step should run before release?
• The city SOC wants AI API anomalies visible alongside other security events. What defensive integration is appropriate?
• Engineers edit production system prompts for the transit assistant without tickets. What control alignment is missing?
• Automated clients hammer the transit bot and spike cloud LLM bills. What defensive engineering focus is needed?
• A water-utility agent can create work orders in the CMMS. What defensive control best reduces unchecked impact?
• An elections FAQ assistant must reduce generative misinformation risk for voters. What defensive combination is most appropriate?
• County IR responds to a suspected AI abuse case on a municipal assistant. What evidence set is specifically important?
• Library service owners complain that chatbot filters block too many legitimate research questions. What defensive engineering tradeoff must be governed?
• Departments call LLMs with inconsistent logging and policy. What architecture improves consistent civic defensive controls?
• Fire/EMS assistant logs must support attack investigation without exposing sensitive resident details broadly. What design goal fits?
• After every major model upgrade, the city wants automated AI security suites re-run before broader release. What practice does that embed?