Customers, residents and staff increasingly want to know one simple thing: am I dealing with a person or with AI? From 2 August 2026 the EU AI Act made that question a legal one for chatbots and AI-generated content reaching people in Europe, and Australian guidance now sets clear expectations too. This guide explains what the rules say, which ones reach South Australian organisations, and practical wording for chatbots, AI voices, images, letters and reports used by SA councils, agencies, not-for-profits and small businesses.
This is general information, not legal advice. It reflects official guidance as at October 2026. Confirm your own position with the sources linked below and your legal adviser.
The short answer for SA organisations
- Australia has no AI labelling law, but existing law still applies. Under the Australian Consumer Law, letting people believe they are talking to a human, or that AI-made images show something real, can be misleading or deceptive conduct.
- National guidance expects disclosure. The National AI Centre’s Guidance for AI Adoption says people should know when they are interacting with AI, and its guidance on being clear about AI-generated content explains how to label it.
- Automated decisions need privacy policy wording from 10 December 2026 for organisations covered by the Privacy Act. See our automated decisions checklist.
- The EU AI Act’s Article 50 can reach you from Adelaide if your chatbot, app or AI content is used by people in the EU.
- Government buyers expect transparency. Commonwealth agencies publish AI transparency statements under the DTA policy, and suppliers are asked how AI is used and disclosed.
Australia: what is expected now
Guidance for AI Adoption: share essential information
The National AI Centre’s Guidance for AI Adoption (October 2025) sets six essential practices. Practice 4, share essential information, says people should know when they are interacting with AI and understand when AI decisions affect them; its own example is a customer using a chatbot, who should know it is not a human specialist. Getting-started actions are to keep an AI register and make disclosure standard practice, especially where AI makes or influences decisions, generates content that can meaningfully affect people, or could be mistaken for a human. Our explainer on the VAISS guardrails and the six essential practices covers the rest.
Being clear about AI-generated content
The National AI Centre’s voluntary guidance describes three transparency mechanisms, used alone or together:
- Labelling: visible text saying content was created or changed by AI. Easiest to apply, but easiest to remove.
- Watermarking: visible or invisible marks embedded in images, video or audio to trace origin.
- Metadata recording: details in the file about who created it and how. Content Credentials are an emerging standard for this.
The level of disclosure should be proportionate to how much AI was involved and how much the content could affect people. Grammar fixes in an email may need nothing; AI-generated health, recruitment, financial, legal or public communications content needs clearer, often combined, methods. Even small AI edits that change meaning, such as “did” to “did not”, deserve care.
Government
The DTA’s Policy for the responsible use of AI in government requires Commonwealth agencies to publish AI transparency statements and keep use case registers. SA public sector agencies follow Guideline 13.1 on generative AI, which expects outputs to be checked and information protected. Councils are not bound by the DTA policy, but residents expect the same honesty, and FOI and State records law apply to AI-assisted work. Read AI for South Australian councils for the council view.
EU AI Act Article 50: the transparency rules
Article 50 has applied since 2 August 2026. The Digital Omnibus on AI (Regulation (EU) 2026/1744) did not delay it, except for one narrow grace period described below. Its duties fall on two roles: providers, who build or supply an AI system, and deployers, who use one in their work.
| Situation | Who must act | What is required |
|---|---|---|
| Chatbots, AI agents, voice bots and avatars talking to people | Provider | Design the system so people are told they are interacting with AI, unless it is obvious from the context |
| AI-generated text, images, audio and video | Provider | Mark outputs in a machine-readable, detectable way. Systems already on the market before 2 August 2026 have until 2 December 2026 |
| Emotion recognition or biometric categorisation | Deployer | Inform the people exposed to it |
| Deepfakes: realistic images, audio or video of real people, places or events | Deployer | Disclose that the content is AI-generated or manipulated, with lighter rules for clearly artistic or satirical work |
| AI-generated text published to inform the public on matters of public interest | Deployer | Disclose it, unless a person has reviewed it and someone holds editorial responsibility |
Information must be clear, distinguishable and accessible, and given at the latest at the first interaction or exposure. The European Commission published a voluntary Code of Practice on Transparency of AI-generated Content on 10 June 2026, with a section for providers on marking and detection and a section for deployers on labelling, plus a set of optional EU icons. Guidelines on Article 50 explain exclusions, such as assistive standard editing and short codes or source code.
Does it apply to an Adelaide business?
Only where there is an EU link: your AI system is placed on the EU market, or its output is used in the EU. A local cafe’s website chatbot serving Adelaide customers is not the target. An SA software company selling an AI-enabled app to European customers, an exporter running an AI sales assistant for EU buyers, or a tourism operator producing AI video for European markets should check carefully. Our AI laws tracker covers the wider Act and other jurisdictions.
Other countries with labelling rules
- China: measures for labelling AI-generated synthetic content have applied since 1 September 2025, requiring visible and embedded labels on certain content for services offered in China.
- South Korea: the AI Basic Act took effect on 22 January 2026 and includes transparency and labelling duties for generative and high-impact AI.
- United States and United Kingdom: no general federal or UK labelling law as at October 2026, but consumer protection and data protection regulators expect people not to be misled, and some US states have specific rules on chatbots, deepfakes and political ads.
Practical disclosure patterns
The patterns below meet the Australian guidance and are a sensible baseline if you have EU users. Adapt them to your context and have them reviewed.
| Use | Suggested approach | Example wording |
|---|---|---|
| Website chatbot or virtual assistant | Say so in the first message and the window title, and offer a human route | “I’m an AI assistant. I can answer common questions or pass you to our team.” |
| AI voice on phone lines or a voice bot | Say it at the start of the call, before collecting details | “You’re speaking with an automated AI assistant. Say ‘staff member’ at any time.” |
| AI-drafted letters, emails or reports a person reviews | Usually no label needed if a named person checks and owns it; keep a record of AI use | Internal register note: “Drafted with Copilot, reviewed by [role].” |
| Fully automated replies or notices | Label them and give a contact for questions or review | “This message was generated automatically. Contact us to speak with a person.” |
| AI images in marketing or social posts | Label realistic images, avoid AI images of real people or places that could mislead, and keep the file metadata | “Image created with AI.” |
| Council or agency public notices and media | Human review and editorial sign-off always; label any AI imagery or synthetic voice | “Illustration generated with AI. Not a photo of the site.” |
| Meeting transcription or summaries | Tell attendees before recording and let them object; check summaries before sharing | “This meeting uses AI transcription. Notes are reviewed before circulation.” |
Make disclosures work for everyone
- Accessible: put the disclosure in text, not just an icon; make sure screen readers announce it; give voice bots a spoken notice. Our WCAG 2.2 guide covers the basics.
- Early: before people share personal information, not buried in terms of use.
- Honest about limits: say what the assistant can and cannot do, and that answers may need checking.
- Consistent: your chatbot notice, privacy policy, collection notices and staff AI policy should say the same thing.
- Escapable: a clear way to reach a person, especially for complaints, hardship or anything affecting rights.
A 30-day plan
- Week 1: list where people meet AI. Chatbots, phone lines, email automation, marketing content, recruitment tools and AI features inside software you already use. Add them to your AI register.
- Week 2: decide the level. For each, rate AI involvement and potential impact, check for any EU, China or Korea users, and choose label, watermark, metadata or none, with reasons.
- Week 3: write and build. Add chatbot and voice notices, label templates and a human hand-off. Ask vendors whether their tools add machine-readable marks or Content Credentials. Build transparency into your chatbot guardrails.
- Week 4: set the rule and train. Add a disclosure rule to your generative AI acceptable use policy, brief staff who publish content, and diarise a six-monthly review.
How EITSAA helps
We help SA organisations find where people meet AI, decide proportionate disclosure, write chatbot, voice and content notices, and build transparency, human hand-off and logging into the assistants we design. We work alongside your lawyers rather than replacing them. Explore AI Governance & Assurance, see AI agents and chatbots, book a short AI transparency review or call 0402 401 584.
If people were not told they were dealing with AI, or an AI answer misled them, treat it as an incident. Our AI incident response playbook explains the steps.
Disclosure is one control among many for a public chatbot. To review the rest of its design, from the knowledge base to the tools it can call, see threat modelling for apps and AI agents.
Sources
- National AI Centre: Guidance for AI adoption, foundations
- DISR: Being clear about AI-generated content
- European Commission: transparency obligations under Article 50
- European Commission: Code of Practice on Transparency of AI-generated Content
- AI Act Service Desk: implementation timeline
- EUR-Lex: Regulation (EU) 2026/1744, Digital Omnibus on AI
As at October 2026. Last reviewed 25 October 2026. Not legal advice.