Risk Register for an AI Feature

Create a risk register for an AI feature that lists what can go wrong, how likely it is and what reduces each risk.

Added October 15, 2026. Written with AI assistance and checked for clarity and placeholders; not reviewed by a lawyer, accountant, clinician or other professional. How these prompts are made

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Create a risk register for this AI feature: [FEATURE_DESCRIPTION]. Users: [USERS]. Data used: [DATA_USED]. Worst outcome we can imagine: [WORST_CASE].

Cover wrong answers, biased outcomes, privacy leaks, misuse, over-reliance, outage, vendor change and prompt injection. For each risk give a plain description, likelihood and impact on a scale from 1 to 5 with the product shown, an early warning sign, a control, an owner placeholder and a review date.

Return a table sorted by score, the three risks to fix first with a concrete action each, and a note on which risks cannot be removed and must be accepted explicitly by a named decision maker.

Also add a column for the residual score after the control is in place, and flag any risk where the control depends on a person remembering to act, since those controls tend to fail under pressure or in busy weeks.

Close with a brief summary paragraph for a manager who has two minutes: the biggest three risks, what is being done, and the decision or resource needed from them this month.

Use a simple scoring pattern so the register is easy to explain. Multiply likelihood by impact, each from 1 to 5. A wrong answer that is likely and moderately harmful might score 4 times 3, which is 12. A privacy leak that is unlikely but severe might score 2 times 5, which is 10. An outage that is possible but low impact might score 3 times 2, which is 6. Sort by score, but also mark any risk with an impact of 5 for attention regardless of its score. For each risk write the control as an action someone can check, such as masking personal data before sending, rather than a vague intention such as being careful. Review the register after every significant change to the feature.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Invented feature: a chatbot answering delivery questions.
Risk 1, wrong delivery date given: likelihood 4, impact 3, score 12. Early sign: complaints quoting dates. Control: answer only from the tracking record. Owner: support lead.
Risk 2, personal data pasted into chat: likelihood 3, impact 4, score 12. Control: warning text and masking of order numbers in logs.
Risk 3, outage: likelihood 3, impact 2, score 6. Control: fallback message and contact form.
Priority: risks 1 and 2 first; risk 3 accepted by the service manager.

• risk