A city SOC briefs the AI board on threats to the 311 GenAI assistant. Which governance-level recognition is most appropriate?
Select an answer to reveal the explanation.
Short Explanation
Prompt injection isn’t sci-fi—it’s a real threat class for the 311 GenAI assistant. Boards need to recognize it, not hand out exploit recipes or pretend CVE scanners cover chatbots.
Full Explanation
When a city SOC briefs the AI board on threats to the 311 GenAI assistant, the appropriate governance-level recognition is to treat prompt injection as a recognized threat class against that assistant. GenAI systems that accept untrusted resident or ticket text can be steered by crafted inputs to ignore instructions, exfiltrate context, or produce unsafe actions. Boards and risk committees need this threat named so it receives owners, controls, and monitoring—alongside traditional application risks.
Assuming chatbots cannot be manipulated through crafted inputs fails by concept because instruction-following models are inherently influenceable by content in the prompt path. Publishing step-by-step exploit recipes for staff “practice” fails as governance practice because distributing weaponized how-tos increases misuse risk and is not required to acknowledge the threat class. Ignoring GenAI threats because traditional CVE scanners cover everything fails because prompt-layer abuse is not fully represented by classic vulnerability scanning of host packages.
Exam caveat: recognizing prompt injection at board level does not require teaching attack recipes; governance needs impact, likelihood framing, and control expectations (input handling, privilege separation, monitoring, human escalation). Operational check: add prompt injection to the 311 GenAI threat register with a control owner, define abuse-case tests for ticket and chat inputs, and alert on anomalous tool-use or policy-bypass patterns.