Before launching a resident-facing bot, the program and security teams run abuse-case tests for prompt injection and agree on remediation owners. What adoption practice is this?

Including security abuse-case testing in civic AI adoption with clear remediation ownership is the correct practice when program and security teams run prompt-injection tests before launching a resident-facing bot. Resident bots are probed; injection and other abuse cases are predictable enough to test lightly before go-live rather than discover via viral misuse. Agreeing on remediation owners ensures findings become fixes with accountability instead of floating risk notes. The practice works because adoption without abuse testing leaves known attack classes unexamined on a public civic interface.

Skipping all abuse testing because the bot only answers FAQs today fails by concept: FAQ scope does not prevent injection attempts that try to exfiltrate prompts, override policies, or extract hidden context. Publishing full exploit playbooks as the training curriculum for clerks fails by concept—staff need safe-use and reporting guidance, not a public cookbook of exploit steps as primary training. Treating prompt injection as solely a marketing problem fails by concept: injection is a security and product-safety issue requiring technical and process remediation, not messaging alone.

Exam caveat: abuse-case testing should be proportionate and coordinated; it is not permission to run unbounded offensive research against production without rules. Keep resident-facing municipal bot framing. Operational check: define a short abuse-case suite (prompt injection, jailbreak attempts, sensitive-data probes); run it pre-launch with program and security; assign remediation owners and due dates; and retest after material prompt or retrieval changes.