What the Digital Duty of Care bill says about AI chatbots, comments, reviews and other user content, and what a business with these features can do now.
Does the bill cover AI chatbots?
New section 25B(1)(h) of the bill text as passed by the House lists, as an online service, “a service that generates material using artificial intelligence and that uses an internet carriage service in the course of doing so”.
The revised explanatory memorandum says this item is there “for the avoidance of doubt”. In other words, the bill makes clear that AI services are online services.
A chatbot on a business website writes its answers, so it generates material. The bill does not exclude business chatbots, and it sets no size threshold. The Minister can exempt services that pose little risk under section 25B(3), but no exemption exists yet. Our article Does the Digital Duty of Care apply to my business? explains who the duty covers.
What risks does the bill link to chatbots?
The memorandum names “chatbots providing inappropriate and harmful responses” as one of the new harms the current Act cannot deal with well. The bill then adds harms that matter for chatbots in particular.
Section 25E lists material and conduct that is harmful to children. Paragraph (da) covers material or conduct that “simulates emotional intimacy and encourages or fosters a child’s dependence on it in place of human relationships”. The memorandum says this paragraph is meant to capture AI chatbots and companions that simulate emotional intimacy and personality.
Section 25D lists seriously harmful material for everyone, for example material that instructs self-harm or suicide. A chatbot that can produce such material in reply to a prompt is a risk that a provider needs to assess.
What controls does the memorandum expect for AI tools?
The memorandum gives a list for a provider of a public-facing generative AI service. It says relevant controls can include:
- Testing before deployment.
- Testing against reasonably foreseeable misuse.
- Safeguards on user prompts and on generated outputs.
- A way for users to report harmful outputs.
- Monitoring after deployment.
- Review of whether the safeguards still work as the service and its use change.
This list is not a checklist in the law. But it shows the kind of controls that the explanatory memorandum has in mind for AI tools. The current Basic Online Safety Expectations already expect that features such as generative AI and recommender systems can be used safely.
Does adding a chatbot need a new risk assessment?
It can. Section 27(3) says a risk assessment must be done before any significant change to the service that can introduce new or additional risks.
The memorandum gives an example. It describes adding a generative AI tool to an existing service, which creates new ways for users to make harmful material and share it at once. That change can require the provider to assess the risks again. Routine software updates and security patches do not, if they do not change the safety of the service.
So the time to assess a chatbot is before you switch it on, not after a problem. Our article on what a risk assessment asks for explains what each assessment must hold.
How does the bill treat comments, reviews and forums?
User content is not new to online safety law. The Online Safety Act 2021 already defines the services that carry it:
- Section 13A defines a relevant electronic service. It includes chat services and instant messaging that let end-users communicate with other end-users.
- Section 14 defines a designated internet service. It includes a service that allows end-users to access material over the internet.
A website where customers post reviews, a forum, or a live chat between members can fall into these categories. The bill makes each of them an online service under section 25B.
Am I responsible for every post a user makes?
No. The memorandum says that individual pieces of harmful content do not by themselves breach the duty. The question is whether the provider identified the reasonably foreseeable risks and kept effective systems in place to reduce them. Harmful content can still be part of the evidence of a wider failure.
The memorandum gives examples of measures. One is to update and enforce your terms of use, so they set standards of behaviour and ban cyber-bullying, abuse and the sharing of intimate images without consent. Others are complaint handling, content moderation and safety by design.
Section 26B adds one limit. The duty does not require action on lawful communications in private solely between consenting adults. The memorandum says large group chats are not private for this purpose.
What about complaints from people who are not users?
Section 27G says a person responsible for a prescribed online service must have complaint and dispute processes available equally to everyone in Australia, whether or not they use the service. The Commissioner decides which classes of service are prescribed. The memorandum explains why: a person who is not a user can still be the subject of content on the service.
Where does the bill stand?
On 19 November 2026, the bill page shows the status “Before Senate”. If the bill passes, the duty starts the day after a 12-month period that begins on the day of Royal Assent.
What can a business with a chatbot or user content do now?
These steps help you prepare:
- List each chatbot, comment area, review feature, forum or chat window you run.
- For each AI tool, note who supplies it and what it can say.
- Test the chatbot with prompts a misuser can try, and keep the results.
- Check that your terms of use set standards of behaviour.
- Give users and non-users a clear way to report harmful content or replies.
To see the chatbot test and other readiness tools we are building, visit our products page. Our guide How Smart Digital Care works shows the path from a free check to complete, dated records. To talk to us about your chatbot, use our contact page.
Sources
Frequently asked questions
Does the bill cover AI chatbots?
Yes. Section 25B(1)(h) lists a service that generates material using artificial intelligence as an online service. The bill does not exclude business chatbots.
Am I responsible for every post a user makes?
No. The memorandum says that individual pieces of harmful content do not by themselves breach the duty. The question is whether you identified the reasonably foreseeable risks and kept effective systems in place to reduce them.
Does adding a chatbot need a new risk assessment?
It can. Section 27(3) asks for a risk assessment before any significant change that can add risk. The memorandum gives adding a generative AI tool to an existing service as an example.
What controls does the memorandum expect for a public AI tool?
Testing before deployment, testing against reasonably foreseeable misuse, safeguards on prompts and outputs, a way to report harmful outputs, monitoring after deployment, and review of the safeguards.