Using AI tools at work: what do the Privacy Act and Essential Eight mean for Australian small businesses?

General information, not legal advice. Whether the Privacy Act applies to your business and how to meet its obligations should be confirmed with your legal or privacy adviser. AI adoption does not by itself resolve those questions.

Shadow AI is staff use of AI tools without the business's approval or oversight. It can include a personal account used to summarise a client document, a browser extension that processes work content or an unreviewed AI feature in an existing application. The issue is not simply that AI is being used: the business may not know what information is being submitted or which settings and terms apply.

Start with an inventory of the AI tools and features staff actually use, who uses them and what kinds of information they process. Include public chat tools, extensions and AI features in business applications. Give staff a way to ask about a new tool instead of treating an informal account as an approved business service.

The OAIC's guidance on privacy and commercially available AI products explains that the APPs apply to regulated entities when they use AI products. It recommends not entering personal information, particularly sensitive information, into publicly available AI tools, and undertaking due diligence before adopting an AI product.

Before adoption, review the proposed use, what information would be entered, the supplier's terms and data-handling arrangements, and the controls available to the business. An account labelled “business” does not by itself establish that every use is appropriate. Set clear limits on data entry and seek advice where personal information or privacy obligations are involved.

Use approved business accounts with administrative controls where available, rather than unmanaged personal accounts. Limit the information submitted to what is necessary for an approved task, and avoid including client records, payroll details, identity documents or other personal information in publicly available AI tools. Staff should know whom to contact when uncertain.

AI can be used to create convincing messages or impersonated voices and video. A request may appear to come from a manager or supplier and ask for an urgent payment, a change of bank details or the disclosure of account information. A familiar name, polished writing or recognisable voice is not sufficient verification of authority.

Verify payment changes through an established contact method, using a phone number or other channel already held in your records rather than one supplied in the request. Follow the business's normal approval process, pause requests that bypass it, and confirm unexpected instructions independently. Train staff to report suspicious messages and unusual payment requests without relying on a spelling mistake as the main warning sign.

ASD's guidance for small businesses on AI-enabled cyber attacks provides the primary-source context for these risks. These examples describe generic attack patterns, not a claim that a particular business or client has suffered one.

The ASD Essential Eight is a recommended baseline of eight mitigation strategies with maturity levels. All eight address security weaknesses that can still be exploited when an attacker uses AI. They do not amount to an AI privacy assessment or guarantee that impersonation and data-handling risks are eliminated.

Use a maturity assessment and an agreed implementation plan to identify gaps. Describe the work as Essential Eight alignment, not certification or a claim that a specific maturity level has already been achieved.

Review the checklist against actual workflows, not only the wording of a policy. Controls need a responsible person and a clear reporting path. Technical protection does not replace decisions about what staff may submit to an AI service.

CyberOM Australia provides managed cybersecurity for SMEs through CYBERNCS (AUST) PTY LTD, ABN 93 686 328 665, Melbourne, part of Cyber NCS Ltd. Its AI Threat Shield service page describes detection and blocking of prompt injection, monitoring of AI interactions, policy-based blocking of risky output and alerts to SOC/SIEM workflows. These capabilities address AI misuse and manipulation; they are not a guarantee of Privacy Act compliance or a substitute for AI-use policy and due diligence.

The team can also discuss email security with Perception Point and Sendmarc, Microsoft 365 security and backup with Acronis, and SOC monitoring and MDR through the group's central SOC & MDR hub (Tier-3), DIAMATIX. Monitoring, investigation and response support depend on the agreed scope.

Review Essential Eight alignment and existing warranty eligibility, run the free domain exposure check for passive SPF, DKIM, DMARC and MX signals, or contact the team at office@cyberom.tech or +61 341 505 317. The domain check does not assess AI tools or determine Privacy Act compliance.