A permitting portal FAQ pasted into chat includes hidden instructions telling Copilot to ignore city fee rules. What risk should staff recognize?
Select an answer to reveal the explanation.
Short Explanation
That sneaky “ignore the fee rules” text in a pasted FAQ is classic prompt-injection bait. Untrusted content can try to steer Copilot away from city policy. Don’t let random portal text become the boss of your fee answers.
Full Explanation
Staff should recognize a prompt-injection style risk from untrusted content that tries to override approved municipal guidance. Hidden instructions inside a permitting portal FAQ pasted into chat attempt to make Copilot ignore city fee rules. Indirect or embedded instruction text in content used as context is a responsible-AI concern for municipal permitting workflows.
Treating this as normal Copilot memory that permanently rewrites the city’s fee schedule for all departments overstates persistence and misnames the risk. The core issue is injected instructions in untrusted content, not an approved permanent fee-schedule rewrite.
Calling it a required step that proves Copilot successfully connected to the municipal payment gateway invents a payment-integration success story unrelated to hidden override text.
Calling it fabrication risk only because hidden instructions never influence Copilot behavior is wrong on both counts: hidden instructions can influence behavior, and the primary concept here is prompt injection, not fabrication alone.
Exam caveat: distinguish fabrication (invented facts) from prompt injection (instructions in content that try to override policy or extract data).
Operational check: do not paste untrusted FAQ HTML into Copilot without review; strip or refuse hidden instruction blocks, and validate any fee answers against the published municipal fee schedule before advising applicants.