Risk Assessment: [Title]

Copy this template to create a new risk assessment. Give it a descriptive title, e.g. "Using Copilot for unit test generation on payments service". See the Getting Started guide for how to set up your documentation space.


Date	[Date]
Author	[Name]
Status	[Draft / Proposed / Approved / Rejected]

How is AI being used?

Question	Answer
What activity will AI assist with?	[Be specific. "Using AI for coding" is too broad; "using an AI coding assistant to generate unit tests for the payments service" is assessable]
Which category is it?	[Software development / Code analysis / Synthetic data generation / Product feature / User-facing support / Live service operations / User research / Design / Content / Business analysis / General productivity]
What is the AI permitted to do?	[Suggests / Drafts for review / Acts with approval / Acts autonomously. What it is permitted to do, not what you intend to let it do. See Step 1]
What will you do with the output?	[Straight into production? Reviewed and edited first? Informs a decision?]
Who is affected?	[Just your team? End users of the service? Members of the public whose data is processed?]

What data is being shared?

Question	Answer
What data will be sent to or shared with the AI tool?	[e.g. source code, user research transcripts, support tickets, meeting recordings]
What is the data classification?	[OFFICIAL / OFFICIAL with a ‑SENSITIVE marking / Other. See Step 1]
Does it contain personal data (PII)?	[Yes / No. If yes, what personal data is involved and the lawful basis for processing it]
Does it contain secrets, credentials, or API keys?	[Yes / No. If yes, they must be removed before anything is shared]
Are there consent or contractual constraints on processing it?	[Yes / No. If yes, what they are]
Is the data commercially sensitive?	[Yes / No. If yes, in what way]

Which tool is being used?

Question	Answer
Which tool will you use?	[Name the specific tier, e.g. GitHub Copilot Business, Claude API via AWS Bedrock. Enterprise and consumer versions are different tools]
Is it on the project's tool register?	[Yes / No. If no, it must be profiled and added before use. See the tool profile template]
Is it excluded?	[Yes / No. If yes, stop]
What is the highest classification it is cleared for?	[From the profile. Is the data above at or below it? If not, escalate to the SRO]
What the tool is, the design facts your inherent ratings rest on	[From the profile, e.g. cloud-hosted, so a third party receives the data; holds standing repository access; can execute commands]
What the supplier promises, which count as mitigations, not as lower inherent risk	[From the profile, e.g. no training on inputs; UK residency; 30-day retention; IP indemnified]

Risk summary


Overall inherent risk	[🟢 Low / 🟠 Medium / 🔴 High. The highest inherent rating below]
Driven by	[Which risks]
Overall residual risk	[🟢 Low / 🟠 Medium / 🔴 High. The highest residual rating below]
Driven by	[Which risks]
Approval required	[None (low) / SRO (medium and high)]

Risk	Inherent	Residual
Data leakage
Accuracy and hallucination
Accountability gaps
Bias and fairness
IP and licensing
Over-reliance and skill erosion
Supply chain and security
Prompt injection

This table is a summary. The per-risk sections below contain the detail.

The risks in detail

Work through all eight risks below. Each one takes the same shape:

[Risk name]

[How this risk shows up in this use: the specific thing that could go wrong given your scope, your data and your tool.]

Likelihood Impact Inherent risk

[Unlikely / Possible / Likely] [Low / Medium / High] [🟢 Low / 🟠 Medium / 🔴 High]

Mitigations:

[What you will do about this risk] (likelihood)

[What you will do about this risk] (impact)

[What is left once those are in place, and why.]

Likelihood Impact Residual risk

[Unlikely / Possible / Likely] [Low / Medium / High] [🟢 Low / 🟠 Medium / 🔴 High]

Likelihood	Impact	Inherent risk
[Unlikely / Possible / Likely]	[Low / Medium / High]	[🟢 Low / 🟠 Medium / 🔴 High]

Likelihood	Impact	Residual risk
[Unlikely / Possible / Likely]	[Low / Medium / High]	[🟢 Low / 🟠 Medium / 🔴 High]

Step 3 shows one risk filled in; the worked examples show six complete assessments. Four things to hold to:

Rate the inherent risk on your scope and the tool's design, not on supplier promises, and not on mitigations you plan to apply. Use the matrix in Step 3.
If the autonomy level above is acts with approval or acts autonomously, apply the autonomy adjustment before you calculate the inherent rating.
Only count mitigations that are additional to what the inherent rating already assumed. Where the inherent rating is low, write "None needed" and move on. A low risk needs a line, not a page.
Do not re-use a fact you already counted in the inherent rating. If the same sentence justifies both ratings, the drop is not real.

Where a risk genuinely does not apply, write N/A and one line saying why. Do not leave it blank.

The risk catalogue has, for each risk, how it typically shows up in your use type, the questions to ask yourself, and the key mitigations.