This article summarises publicly available guidance from regulators and official sources. It is general educational information only and does not constitute legal or professional advice. Requirements vary by jurisdiction. Consult your regional authority or a qualified professional for advice specific to your situation.
Every AI tool your business uses carries some level of risk. The risk is not always significant, but it is always worth assessing before a tool goes into regular use rather than after a problem occurs. This checklist gives you a structured way to evaluate any AI tool across the five dimensions that matter most for Australian SMBs: data privacy, accuracy and reliability, security, accountability, and bias. Complete one assessment per tool. The output is a traffic-light rating that tells you whether the tool is ready to use, whether conditions or safeguards are needed, or whether it needs more investigation before use.
In short: Work through each dimension and rate each question Green (satisfactory), Amber (some concern, manageable), or Red (significant concern or unknown). A mostly Green result means the tool is likely suitable for the intended use with standard controls. More than two Reds in any dimension means the tool warrants further investigation before deployment. A Red on any question in Dimension 1 (Data Privacy) should be resolved before the tool handles personal information. The Australian Guardrails referenced are from the federal Department of Industry, Science and Resources' Voluntary AI Safety Standard (DISR Standard); references are for orientation, not compliance verification.
This checklist works best when you are assessing a specific AI tool for a specific use case in your business. Fill in the tool name and intended use at the top before starting so the assessment stays focused. A general assessment of a tool without reference to how you plan to use it will produce a less useful result.
Tool Details
- Tool being assessed: ______________________
- Intended use in this business: ______________________
- Type of data this tool will process: ______________________
- Date of assessment: ______________________
- Assessed by: ______________________
Dimension 1: Data Privacy
Data privacy risk is the most frequently overlooked dimension in SMB AI assessments, and the one most likely to create regulatory exposure. The core questions are: what data will this tool process, where does that data go, and does that match what your business is permitted to do with that data under the Privacy Act and applicable sector requirements. Australian Privacy Principle 8 (APP 8) requires that cross-border disclosure of personal information is managed appropriately. Most AI tools send data to overseas servers.
Picture the office manager at that 22-clinician physiotherapy group again: one of the physios wants to start using an AI tool that listens in on consultations and drafts the clinical notes automatically. Before running this checklist, her only real option was to trust the vendor's website and either approve the tool on faith or block it out of caution. After working through Dimension 1, she has a clear conditional answer instead of a guess: approve for trial, but only once the vendor confirms in writing where consultation audio is stored and the practice updates its Privacy Act collection notice to cover it. That is a decision she can defend to a patient, a regulator, or the practice's insurer.
Data Privacy Questions
- Have you identified what types of information this tool will process? (personal information, sensitive information, health information, financial information, internal business data only)
- Does the tool's privacy policy or data processing terms clearly state where data is stored and processed?
- If data is stored or processed outside Australia, has the business assessed whether this is consistent with APP 8 cross-border disclosure obligations?
- Does the vendor's terms prohibit using your data inputs to train their models, or do they offer an opt-out? (On free and lower-tier plans, many AI vendors do use inputs for training.)
- Does the vendor have a clear data retention policy? Do they state how long your data is retained and whether it can be deleted on request?
- Is there a Data Processing Agreement (DPA) or equivalent available from the vendor for businesses in regulated industries or those handling sensitive personal information?
- Have the data handling terms been reviewed against your existing privacy policy and any sector-specific obligations that apply to your business?
Red flag conditions: No clear data handling policy available; data used for model training without opt-out; no data deletion mechanism; unclear whether data leaves Australia with no APP 8 assessment completed. For tools that will process personal information about clients or customers, a Red on any item above warrants resolution before deployment. The OAIC's guidance on using commercially available AI products is at oaic.gov.au.
Dimension 2: Accuracy and Reliability
AI tools can produce confident-sounding outputs that are factually wrong. The risk from inaccurate outputs depends entirely on how the output will be used. An AI that summarises a meeting note that a human then reviews carries low accuracy risk. An AI that generates a quote, a contract clause, or a clinical note that goes to a client without review carries much higher risk. The accuracy questions focus on the consequences of an error in the specific use case you have in mind.
Accuracy and Reliability Questions
- Have you tested this tool on representative tasks from your business with outputs you could verify independently? (Do not rely solely on vendor-provided benchmarks.)
- Is there a human review step before AI outputs are used for decisions, sent to clients, or incorporated into documents that have legal or commercial significance?
- Does the tool's use case in your business involve areas where hallucinations or errors would have significant consequences? (Legal documents, clinical notes, financial advice, safety instructions are high-consequence. Internal drafts for human review are lower consequence.)
- Does the vendor clearly document the tool's limitations, known failure modes, and the types of tasks it should not be used for?
- Is the tool's performance consistent across the language, domain, and context of your business? (Many AI tools perform well in general English but less well on Australian-specific regulatory content, industry jargon, or technical terminology.)
- Is there a process for staff to flag and escalate AI errors when they are found?
Red flag conditions: Tool will produce outputs that go to clients or inform significant decisions without human review; tool has not been tested in your specific domain; error reporting process does not exist. For high-consequence use cases (legal, clinical, financial), Amber ratings here warrant explicit verification protocols rather than just a general review step. The DISR Voluntary AI Safety Standard Guardrail 5 addresses human oversight and review; see industry.gov.au for the full standard.
Dimension 3: Security
AI tools are software, and the same basic security questions that apply to any business software apply to AI tools. The questions in this dimension cover access control, vendor security posture, and the specific risks that arise from AI tools storing or processing sensitive data in vendor-hosted environments.
Security Questions
- Does the vendor hold recognised security certifications appropriate for the data sensitivity of the tool's use case? (SOC 2 Type II and ISO 27001 are common benchmarks. Absence of certifications does not automatically disqualify a tool but warrants scrutiny for high-sensitivity uses.)
- Can access to the tool be managed through your business's existing identity management? (Single sign-on, multi-factor authentication, and role-based access reduce the risk of unauthorised access.)
- Is the vendor's breach notification policy clear? Would they notify you promptly if your business data was involved in a security incident?
- Does the tool require more access permissions than it needs for the intended use? (An AI writing tool that requests access to your entire file system or email history when it only needs to draft documents is requesting excessive permissions.)
- Is there an audit log or activity history available so your business can review how the tool has been used and by whom?
- Is there a clear process to offboard from the tool and delete business data from the vendor's systems if you stop using it?
Red flag conditions: No security certification for a tool that will process sensitive data; no breach notification commitment; tool requests excessive permissions; no data deletion on offboarding. The cyber.gov.au AI security guidance at cyber.gov.au covers baseline security practices for businesses using AI tools, including vendor assessment questions.
Dimension 4: Accountability
Accountability means being able to explain, justify, and if necessary correct the decisions and outputs your business produces with AI assistance. This dimension is about whether your business maintains genuine oversight of AI use or whether AI outputs are being used as a black box. Guardrail 7 of the DISR Voluntary AI Safety Standard addresses contestability and human ability to override AI outputs; this dimension reflects those principles in a practical SMB context.
Accountability Questions
- If an AI output were challenged by a client, regulator, or court, could your business explain what tool produced it, what inputs were used, and why the output was accepted as reliable?
- Is there a clear policy for staff on when they can act directly on AI output versus when they must apply human judgement before acting?
- If the business receives a complaint about an outcome that involved AI, does a person with authority to investigate and respond exist?
- Does the business keep sufficient records of significant AI-assisted decisions to reconstruct what happened if needed? (This does not require logging every AI interaction, but decisions with significant consequences for clients or third parties should be documentable.)
- Is the tool's behaviour consistent and explainable enough that staff can give a reasonable account of why an output was produced? (Opaque black-box outputs that staff cannot interpret or challenge create accountability gaps.)
- Does the business have a process for correcting errors if an AI output was used and later found to be wrong?
Red flag conditions: Staff cannot explain AI outputs to clients or regulators; no process for correction when errors are found; no named person accountable for AI-assisted decisions. The DISR Voluntary AI Safety Standard is at industry.gov.au/ai-safety-standard.
Dimension 5: Bias and Fairness
AI tools can produce outputs that systematically disadvantage certain groups of people based on characteristics like gender, age, ethnicity, or disability, even when no such intent exists. This is particularly relevant when AI is used in processes that affect people's opportunities or access to services: hiring, lending, tenant screening, insurance, and benefits assessment. Australian anti-discrimination law and the upcoming automated decision-making provisions of the Privacy Act both have relevance here.
Bias and Fairness Questions
- Will this tool be used in any process that makes or informs decisions about individual people (hiring, screening, pricing, lending, tenancy, access to services)?
- If yes, has the business reviewed the vendor's documentation on how the tool was trained, what datasets were used, and any known bias-related limitations?
- Is there a human review step for any AI-assisted decision that affects an individual's access to employment, tenancy, credit, or other significant outcomes?
- Does the business monitor the outputs of this tool over time to identify any patterns that may suggest systematic differences in how different groups of people are being treated?
- Are the people whose opportunities may be affected by AI-assisted decisions able to seek a human review of a decision if they request it?
- Is the business aware of its obligations under applicable anti-discrimination law (federal and state/territory) that apply independently of AI governance requirements?
Red flag conditions: Tool used in people-affecting decisions without human review; no vendor documentation on training data or known biases; no appeals process for affected individuals. For tools used in hiring, lending, or tenancy decisions, Reds here carry significant legal and reputational risk. The Australian Human Rights Commission has published guidance on AI and human rights at humanrights.gov.au.
Reading Your Assessment
Once you have completed all five dimensions, apply the following overall rating:
Overall Rating
- Green: Deploy with standard controls. Mostly Green across all dimensions, with no more than 1-2 Ambers in any dimension and no Reds on data privacy. Standard controls mean a use policy, a human review step for consequential outputs, and a point of contact for issues.
- Amber: Deploy with conditions. Several Ambers across dimensions or Reds in dimensions other than Data Privacy. Define specific conditions to address the Amber and Red items before or during deployment. Document the conditions and review them after 90 days of use.
- Red: Further investigation required. One or more Reds in Data Privacy, or Reds in multiple other dimensions, or a use case where accuracy or bias concerns are significant and unsatisfied. Do not deploy until the Red items are addressed. The specific items flagged indicate what investigation is needed.
For most off-the-shelf AI tools used for internal business tasks (drafting, summarising, scheduling), a completed assessment will typically produce mostly Greens and Ambers. The value of the assessment is not that most tools fail it, but that it surfaces the specific gaps that are worth addressing and documents that the business applied due diligence before deployment.
Methodology (Real-World, Verified)
This guide is researched against primary regulatory sources and official regulator guidance, verified as of the date shown, and written for a business with no dedicated compliance function.
Read our full methodology and independence and disclosure policy.
Related reading: our can staff upload customer data to AI tools, our AI and the Privacy Act guide, and our free AI staff policy template.
Related reading: our Claude AI review for Australian business.
Try our free AI Privacy Risk Scorer to score your current AI tool setup against Privacy Act requirements.
Try our free AI Compliance Checker to check whether your AI tools meet your compliance obligations.
Related reading: our free AI acceptable use policy template and our AI governance by region.
Does every AI tool I use need this assessment?
Not necessarily to the same depth. A tiered approach works well for most businesses. Tools that will process personal information about clients or customers, or produce outputs used in consequential decisions, warrant the full five-dimension assessment. Tools used only for internal tasks that do not involve personal information (drafting internal emails, generating internal meeting summaries, internal process automation) can be assessed more lightly, focusing mainly on the security and data privacy dimensions. A useful rule of thumb: if the tool touches personal information, run the full assessment. If it does not, a lighter security and data handling check is usually sufficient.
What is the DISR Voluntary AI Safety Standard and does it apply to my business?
The DISR Voluntary AI Safety Standard is a set of ten guardrails published by the Australian Government's Department of Industry, Science and Resources. It is voluntary, meaning there is no legal requirement for SMBs to comply with it. However, it represents the government's view of what responsible AI practice looks like and is increasingly referenced by regulators and procurement bodies. The guardrails most relevant to this checklist are Guardrail 3 (human oversight of AI systems) and Guardrail 6 (contestability: the ability for people affected by AI decisions to challenge them). The full standard is at industry.gov.au/ai-safety-standard.
What is the difference between this checklist and a vendor due diligence checklist?
A vendor due diligence checklist assesses the AI vendor as a business partner: their security certifications, contract terms, breach notification, data retention, and commercial commitments. This per-tool risk assessment focuses on the specific use case risk: what data does this tool process, how accurate is it for this task, and who is accountable for the outputs. Both assessments are useful and complementary. For a new AI vendor relationship, run both. For a well-established vendor where a vendor due diligence assessment has already been done, this per-tool checklist focuses the assessment on the specific deployment. Our AI vendor contracts guide covers the due diligence angle.
How often should I reassess a tool once it is deployed?
A review at 90 days after deployment is a practical starting point, to assess whether the tool is being used as intended and whether any issues have emerged. After that, the triggers for a reassessment include: a significant update to the tool's terms of service or privacy policy, a change in how the tool is being used in your business (new use case, new type of data, higher volume), a security incident at the vendor, or a regulatory change relevant to your industry. For tools in high-sensitivity use cases, an annual review is reasonable even in the absence of specific triggers.
Can I use this checklist to assess AI features built into tools I already use?
Yes, and this is an important use of the checklist that many businesses overlook. AI features built into existing tools like Microsoft 365 Copilot, Xero AI, HubSpot AI, or your accounting software all warrant the same assessment as standalone AI tools, because they process business data and produce outputs that affect business decisions. In some cases, AI features in established business tools have more transparent data handling documentation than standalone AI apps, but they still need to be assessed for the specific use case and data types involved in your business.
The information in this article is general in nature. It reflects a summary of publicly available guidance and does not constitute legal, privacy, or professional advice. Your obligations will depend on your specific situation, jurisdiction, and business circumstances. Do not rely on this article as a substitute for qualified legal or professional advice.
Before assessing specific tools, check whether your business has the governance foundations in place. Our AI readiness checklist covers the four dimensions that consistently determine whether AI adoption goes well or poorly: governance, technical, risk, and people.
Check Your AI Readiness