AI Application Architecture and Development Frameworks

GAIPS · 25 questions

• A city GenAI stack uses an orchestration framework whose plugins can call arbitrary HTTP tools. What should security require first for those plugins?
• A permitting app wires frontend to API gateway to LLM to RAG with no dedicated policy enforcement point. Where should security place authorization and content-policy checks in that GenAI architecture?
• County IT pins an old vulnerable version of a popular LLM SDK in the 311 service. What GenAI application-architecture risk does that primarily create?
• A housing chatbot trusts client-supplied “system” role messages sent from the browser. What architecture hardening is most appropriate?
• Transit builds a LangChain-style agent chain and security reviews callback URLs embedded in tool configs. Which integration-layer risk should they prioritize?
• Library services pull a third-party “municipal AI starter kit” from an unverified registry. What diligence step is most important before adopting it?
• A public-health app stores LLM provider API keys in the frontend JavaScript bundle. What architectural fix best protects those credentials?
• A court e-filing assistant streams tokens to the UI without output filtering middleware. What should architecture add before citizen display?
• A school-district architecture diagram omits the logging and analytics sidecar that receives full student prompts. Why does that matter for GenAI security?
• A water-utility GenAI microservice uses one shared service account across development, test, and production. What identity architecture change is most important?
• A city open-data GenAI API lacks rate limits, allowing scrapers to burn budget and probe jailbreaks. What API-design control should architecture add?
• Emergency operations integrates an SMS gateway as a GenAI tool without signing outbound messages. What integration security gap should be closed?
• Appsec reviews prompt-template repositories across municipal repos and finds hard-coded secrets. What hygiene practice should GenAI architecture enforce?
• A parks mobile app embeds a WebView that posts prompts to an undocumented staging LLM endpoint. What architecture problem should security raise?
• A county benefits bot uses multi-tenant SaaS LLM features without reviewing tenant data isolation guarantees. What should architecture evaluation demand?
• 311 architecture allows the model to write directly to the case-management database through ORM tools. How should write paths be constrained?
• Security finds unsigned model-serving containers being pulled into the city’s Kubernetes cluster. What supply-chain control should architecture require?
• A housing authority GenAI BFF aggregates tools from three vendors with inconsistent authentication schemes. What integration improvement is most important?
• A clerk workflow uses an AI coding assistant to generate production handlers that ship with unsanitized SQL. What supply-chain or quality risk should review gates address?
• A transit GenAI architecture diagram shows “trust the model” as the authorization layer for fare-dispute decisions. What correction should security insist on?
• A city CISO requires an AI SBOM listing models, embeddings, frameworks, and tool plugins for the civic GenAI platform. What does that inventory primarily enable?
• A public-records portal’s GenAI feature shares session cookies with a third-party widget. What integration privacy risk should architecture eliminate?
• County IT evaluates synchronous versus queued GenAI call patterns for a resident assistant. Which benefit do queued patterns most directly support?
• A permitting microservice conflates retrieval, generation, and payment APIs under one process identity. What separation should architecture apply?
• Appsec reviews whether a framework’s default memory and chat-history store encrypts resident conversations at rest. What control is being validated?