An AI security architect must review threat models before any municipal chatbot goes public. What accountability does this role provide?

AI security architect approval based on threat-model review before public release is the accountability this role provides for municipal chatbots. The architect ensures abuse cases, data flows, trust boundaries, integration risks, and mitigations are examined before a city chatbot faces residents. That gate works because public civic bots expand the attack and misuse surface (prompt injection, data exfiltration via retrieval, social engineering at city scale), and architecture-level review catches design flaws that late testing alone may miss.

AI security architect approval based on penetration-test results before public release fails because penetration testing finds late implementation flaws, not architecture-level abuse cases, data flows, trust boundaries, and mitigations—the stem requires a threat-model review, not a pen-test gate. Municipal CIO approval based on formal threat-model review before public release fails accountability because CIO sign-off is the wrong role—the stem requires the AI security architect's accountable pre-release gate, not executive approval. AI security architect approval based on threat-model review before design freeze fails timing because a design-freeze approval does not gate public release—the stem requires review before public release, after tools, data sources, and trust boundaries have settled.

Exam caveat: a signed threat model that is never updated after major retrieval corpus or tool-plugin changes becomes stale; approval should be tied to meaningful design deltas. Operational check: require AI security architect sign-off as a release gate for public municipal chatbots, store threat-model artifacts with the model card, and re-open review when tools, data sources, or trust boundaries change.