System under assessment
- AI system / tool
- Purpose and task
- Who is affected by its output
- e.g. Internal staff only / customers / job applicants
AI Risk Assessment Template
AI risk assessments often reduce to a data protection review, which misses the failure modes specific to these systems: confidently wrong output, drift, prompt injection, and decisions affecting individuals. This template covers those explicitly.
What you get
- Risk dimensions covering data, accuracy, bias, security, and legal exposure
- Prompt injection and tool-access risks for AI with system integrations
- A human oversight section proportionate to the decision's consequence
- A scored summary with a go / conditional / no-go outcome
How to use this template
- 1
Scale the assessment to the consequence. An AI drafting internal meeting notes and an AI screening job applicants need very different scrutiny. Decide the consequence tier first, and assess proportionately.
- 2
Assess what happens when it is confidently wrong. The characteristic AI failure is not an error message. It is a fluent, plausible, incorrect output that a human accepts. Ask who would catch it, how, and what the consequence is if nobody does.
- 3
Treat system access as a separate risk. An AI that can read your documents or call your APIs is an identity with permissions. Prompt injection via untrusted content can turn it into a confused deputy acting on an attacker's behalf.
What's inside
Here's a preview. Unlock the free download to get all 3 sections (2 more below).
Related templates
More free templates IT teams use alongside this one.
AI Governance
Get templateAI Tool Inventory
You cannot govern AI usage you cannot see, and most AI adoption happens through free tiers and browser extensions that never touch procurement. This inventory is built to surface those, which a record of what you bought will not show.
AI Governance
Get templateAI Incident Log
AI problems rarely arrive as incidents. They surface as a wrong figure in a report or a customer noticing something odd, and are fixed quietly without anyone recording a pattern. This log makes the pattern visible.
AI Governance
Get templateAI Acceptable Use Policy
Banning AI tools does not work. It moves usage onto personal accounts where you have no visibility at all. This policy assumes people will use AI and makes the boundary explicit: which tools, which data, and how to get something new approved.
Frequently asked questions
What should an AI risk assessment cover?
+
Data exposure and training terms, accuracy and the consequence of confidently wrong output, bias where people are affected, security including prompt injection where the system has tool access, legal and IP exposure, vendor dependency, and the human oversight arrangement.
What is prompt injection?
+
An attack where instructions hidden in content the AI processes (a document, a web page, an email) cause it to act against the user's intent. It is most serious where the AI has access to systems or data, because it can then act as a confused deputy.
When does AI need extra scrutiny under regulation?
+
Generally when it makes or materially influences decisions about people (employment, credit, access to services) or affects safety. The EU AI Act and similar frameworks tie obligations to use case and consequence rather than to the technology itself.