An AI risk assessment template gives a business a simple way to pause before a useful new tool becomes an unmanaged part of daily work. It does not need to be a lengthy audit. It is a short, repeatable review that asks what the tool will do, what information it will handle, what it can connect to, and what could go wrong if its output is trusted too quickly.
That pause matters because AI features often arrive inside tools a team already uses. A meeting platform offers automatic summaries. A customer system offers drafting help. A file-sharing platform offers search over documents. Someone may only be trying to save time, but the new feature can change where business information goes, who can access it, and how a decision gets made.
The practical goal is not to reject every new idea. It is to separate low-risk experimentation from uses that need a closer look. The NIST AI Risk Management Framework groups responsible AI work around governance, context, measurement, and management. For a small business, that can become a direct set of questions people can answer before the tool is approved.
When to use this AI risk assessment template
Use this template whenever a team wants to introduce an AI tool, enable a new AI feature, connect an AI service to a company account, or move a pilot into normal operations. It is especially important when a tool could see customer information, employee information, financial details, contracts, internal files, passwords, security information, or regulated data.
A light review may be enough for a public, low-stakes use. For example, an approved business account might help an employee generate ideas for a public event description. A more careful review is needed when the same tool will summarize client meetings, analyze uploaded documents, answer customer questions, or make recommendations that affect money, access, people, or compliance.
The difference is not the brand name on the tool. It is the business context. Two teams can use the same AI product in very different ways. One may use it for generic drafting. The other may connect it to confidential files. A useful assessment captures that difference before the second use becomes routine.
AI risk assessment template
Use the following questions as a working record. A short document or shared form is enough. The value comes from getting clear answers, assigning an owner, and recording the conditions for approval.
- Use and owner: What job will the tool help with, which team will use it, and who is accountable for that use?
- Information involved: What will employees type, upload, copy, summarize, or connect? Classify it as public, internal, confidential, or restricted.
- Accounts and connections: Will the tool connect to email, shared drives, customer systems, meeting recordings, cloud storage, or other business accounts? Which permissions are required?
- Vendor safeguards: Does the business use an approved account? Can the vendor use submitted data to train its models? Can an administrator manage users, review activity, and remove access?
- Output and review: What does the tool produce, who checks it, and what must never be sent, approved, or acted on without human review?
- Risk level and decision: Is the use approved, approved with safeguards, limited to a pilot, or not approved? Record the reason and the date for review.
A template should make a decision easier, not create a false sense of certainty. If an answer is unclear, treat that as a reason to ask more questions before connecting the tool to business information. A vendor that cannot explain how data, access, and retention work may not be a suitable choice for a sensitive workflow.
Map the information before sharing it
The first serious question is what information the tool will receive. Teams sometimes focus only on the prompt they type, but connected features can also retrieve files, calendar details, email, recordings, or customer records. The assessment should cover both direct uploads and background access.
A simple four-level model is often enough. Public information is already intended for anyone to see. Internal information is business material that is not public but would not normally cause material harm if seen by the wrong person. Confidential information includes client records, business plans, private contracts, and financial information. Restricted information includes passwords, access tokens, security incident details, health information, and data with specific legal or contractual obligations.
As a default, keep confidential and restricted information out of unreviewed AI tools. If a business wants to use such information, the assessment should document why, which business account is required, what contract or settings protect the data, and what limits apply. This follows the same discipline used for cybersecurity and secure file sharing: understand the information before deciding where it can go.
Review access, connections, and vendor terms
An AI tool can be risky without ever receiving a copied-and-pasted prompt. A meeting assistant may read calendar details. A connected search feature may browse shared drives. A browser extension may see pages employees open. The assessment should list every system the tool can access and the permissions it needs.
Use company-managed accounts whenever the tool will be part of business work. Personal accounts make it harder to manage access, apply settings, review activity, or remove a former employee. Confirm whether administrators can limit data sharing, turn off training on submitted content where that option exists, and control which integrations are enabled.
The OWASP guidance on risks in AI applications is a useful reminder that AI systems can introduce risks beyond ordinary software, including sensitive information disclosure, excessive permissions, and unsafe output handling. A small business does not need to become an AI laboratory. It does need to know which business systems the tool can reach and whether that access is truly necessary.
Decide where human review is mandatory
AI can help draft, summarize, sort, and suggest. It should not quietly become the final decision-maker for work that affects people, money, legal commitments, customer trust, or security. The assessment should identify the person who checks the output and what they must verify before the business acts.
For a public marketing draft, review may mean checking facts, tone, and the final approval process that already exists. For a customer reply, it may mean verifying account details and commitments. For a security use, it may mean preventing the tool from receiving passwords, keys, system diagrams, or incident evidence unless an approved process explicitly allows it.
Set clear boundaries. A tool may prepare a first draft but not send messages on its own. It may summarize a meeting but not create a personnel record without review. It may flag a possible issue but not suspend an account, release a payment, or change a configuration without a qualified person confirming the action.
Use a simple decision scale
Not every tool needs the same approval path. A decision scale keeps the process proportional and stops low-risk ideas from getting lost in unnecessary delay.
- Low risk, approved: The use involves public or low-sensitivity information, no sensitive integrations, and a clear human review step.
- Moderate risk, approved with safeguards: The tool uses internal information or limited integrations, so the business requires a managed account, restricted permissions, defined review, and a scheduled reassessment.
- High risk, pilot only: The tool touches confidential information, affects an important workflow, or needs broad access. Limit the pilot, document controls, and approve expansion only after review.
- Not approved: The vendor cannot provide enough clarity, the permissions are too broad, the data use is not acceptable, or the business cannot put a meaningful review process in place.
Record the decision in plain language. An owner should be able to see why the tool was approved, what restrictions apply, and when the decision will be reviewed again. This helps a business avoid the common problem of a useful trial becoming permanent without anyone revisiting its access or data practices.
Turn the template into an operating habit
Keep the completed assessment with other vendor and technology records. Add it to the process for buying new software, enabling major AI features, or connecting a new service to a business account. Ask teams to raise a short request before a new tool receives company data, rather than after a connection has already been made.
Pair the template with a clear AI acceptable use policy. The assessment decides whether a specific tool and use case are suitable. The policy gives every employee the everyday rules for approved tools, protected information, human review, and reporting mistakes. Together, they make safe behavior easier to repeat.
Review approved uses at least annually, and sooner when a vendor changes its terms, adds an integration, begins handling a new type of information, or the team expands the workflow. Businesses should also have a simple route for reporting accidental sharing, unexpected tool behavior, or an unapproved connection. That is the same practical preparation behind dependable managed security and IT services.
Keep a decision record the next person can use
A completed assessment should leave behind a useful record, not just a checkmark in a purchasing request. Record the approved use, the tool account, the business owner, the data limits, the connections allowed, the human review step, and the date the decision should be revisited. That way, a new manager or technology partner can understand the guardrails without trying to reconstruct an old conversation.
For example, an approval might state that a meeting assistant may summarize internal project meetings through the company account, but may not record calls with clients, access shared legal folders, or send a summary without the meeting owner reviewing it. Those limits turn a vague approval into an instruction people can follow. They also make it easier to notice when a new feature or workflow falls outside the original decision.
Keep the record close to the vendor information, account ownership, and other technology decisions the business already maintains. When an employee leaves, a vendor changes its product, or a team wants to extend a pilot, the assessment gives the reviewer a clear starting point. This is particularly valuable for small teams, where a few people often carry institutional knowledge that would otherwise live only in email threads or memory.
Finally, make the decision visible to the people who need it. The person using the tool should know the purpose and boundaries. The accountable leader should know when review is due. The technology team should know which account and connections are allowed. Clear ownership makes the template a practical safeguard, not another document that disappears after approval.
How H3Systems helps businesses assess AI risk
H3Systems helps businesses move from informal AI experimentation to clear, workable controls. That starts with understanding where teams already use AI, which information and accounts are involved, and which workflows deserve the closest review. From there, H3Systems can help businesses create approval questions, limit access, set practical data rules, and establish the human review steps that fit the work.
The aim is not to slow down useful adoption. It is to give owners a reliable way to say yes with conditions, run a limited pilot, or decline a tool that creates more exposure than value. Businesses that want help applying this template to their own systems can request an AI risk review with H3Systems.
