A team wants to launch a customer-facing LLM feature. As the accountable executive, what do you require before it ships?
Require AI governance in proportion to risk. Apply risk tiering to what the output does, who sees it and whether errors reverse. Every tier needs a named owner, a tested kill switch and fallback, data rules and monitoring; higher tiers add pre-agreed evals, adversarial testing, human review and legal sign-off.
Answer
Short answer
Require AI governance in proportion to the risk, not one checklist for every feature. Start with risk tiering: does its output inform, recommend or act; who sees it; and can a wrong answer be undone? Every tier needs a named owner, a tested kill switch with a non-AI fallback, data-handling rules and production monitoring. Higher tiers add an evaluation with pass criteria agreed in advance, adversarial testing, measured human review, a staged rollout and a written risk acceptance that legal and compliance have seen.
Was this answer helpful?
Likely follow-ups
- The team says a kill switch would take a sprint to build. Do you delay the launch for it?
- Product wants to move a feature from recommending to acting on the customer's behalf. What changes in your requirements?
- Your provider retires the model the feature was approved on. Does the feature need to be approved again?
- Who signs the written risk acceptance when the business owner and the engineering owner disagree?
Related questions
- Two departments share ownership of a critical service and blame each other after every incident. How do you fix it?hardAlso on accountability5 min
- A wrong answer from the model would cost real money. How do you put it in the product anyway?hardAlso on evaluation5 min
- Your provider announces the model you depend on retires in six weeks. Walk me through what you do.hardAlso on evaluation6 min
- How would you test a model for bias?hardAlso on evaluation6 min