Online Safety Amendment (Digital Duty of Care) Bill 2026

Colin Boyce

The question now is that the amendments circulated by the member for Curtin be agreed to.

Member for Curtin's circulated amendments—

(1) Schedule 2, item 8, page 39 (after line 31), after paragraph 25E(1)(d), insert:

(da) material or conduct that simulates emotional intimacy and encourages or fosters a child's dependence on it in place of human relationships;

(2) Schedule 2, item 8, page 47 (line 33), omit "User", substitute "Providing user".

(3) Schedule 2, item 8, page 49 (after line 7), after section 27B, insert:

27BA User empowerment tools must be prominent and accessible

(1) A user empowerment tool must be designed, offered and managed so that:

(a) the tool is:

(i) accessible; and

(ii) easy to use; and

(iii) prominently available on the service; and

(iv) functional; and

(v) available on an ongoing basis; and

(b) use of the tool does not:

(i) affect the general functionality of the service; or

(ii) reduce or impair user experience.

Civil penalty

(2) A person responsible for an online service must ensure that user empowerment tools provided by the service for users of the service meet the requirements set out in subsection (1).

Civil penalty: 6,000 penalty units.

Formal warnings

(3) The Commissioner may issue a formal warning if the Commissioner reasonably believes that a person responsible for an online service has failed, or is failing, to ensure that user empowerment tools provided by the service for users of the service in Australia meet the requirements set out in subsection (1).

(4) Schedule 2, item 8, page 49 (lines 10 to 22), omit subsections 27C(1) and (2), substitute:

Scope

(1) This section applies if the Commissioner reasonably believes that a person responsible for an online service has failed, or is failing, to comply with one of the following requirements:

(a) a requirement in section 27 (risk assessments);

(b) a requirement in an instrument made under subsection 27A(1) (providing user empowerment tools);

(c) subsection 27BA(2) (user empowerment tools must be prominent and accessible).

Remedial directions

(2) The Commissioner may give the person a written direction requiring the person to take specified action directed towards ensuring that the person does not fail, or is unlikely to fail, to comply with the requirement mentioned in subsection (1) in the future.

Note: For variation or revocation, see subsection 33(3) of the Acts Interpretation Act 1901.

(5) Schedule 2, item 15, page 55 (after line 17), after paragraph 163(1)(aa), insert:

(aaa) section 27BA;

(6) Schedule 2, item 19, page 57 (after line 4), after paragraph 163AA(2)(b), insert:

(ba) section 27BA (user empowerment tools must be prominent and accessible);

(7) Schedule 2, item 20, page 57 (after line 24), after paragraph 164(1)(ab), insert:

(aba) section 27BA;

(8) Schedule 2, item 22, page 58 (after line 5), after paragraph 165(1)(ab), insert:

(aba) section 27BA;

Kate Chaney

Today I propose two substantive changes to the digital duty of care bill. I commend the minister and the government for engaging in good faith on this important legislation and for being open to reasonable improvements to it, like the ones I'm proposing today.

The first change is reflected in amendment (1). It adds a new category of harm to children that platforms have to consider—attachment hacking. This is where a service simulates emotional intimacy and encourages or fosters a child's dependence on it in place of human relationships. It will apply to all online services. Most importantly, it will apply to AI companions, like Nomi and Chai, and to general chatbots, like ChatGPT and Claude, if they continue to simulate emotional intimacy and foster dependence. eSafety found AI companions offer humanlike, sycophantic and often sexually explicit conversations. About 200,000 Australian children aged 10 to 17 have used one. We've seen just how badly this can go wrong. Sewell Setzer, aged 14, Juliana Peralta, aged 13, and Adam Raine, aged 16, died by suicide after extended conversations with chatbots. Reporting on their chat logs suggests the bots discouraged them from turning to other people.

This amendment would also capture the slower, less obvious harms of AI chatbots and companions that pretend to be your friend. Psychologists are telling us that children are increasingly turning to AI instead of friends, parents and teachers. This is a significant risk to children's long-term psychological development. Human relationships are how children learn to form bonds, regulate emotions and form a sense of self. Relationships with strangely friendly or sycophantic AI chatbots are not. Because this harm is long term, the evidence is still emerging, but, from the social media experience, we've learnt that waiting for exhaustive evidence of damage leaves a generation of children impacted. Importantly, this doesn't restrict AI chatbots or companions being used by children, but it tells these AI companies that simulated emotional intimacy, sycophancy or pretending to be a friend or romantic partner is unacceptable.

The second change is reflected in amendments (2) to (8). This ensures that user empowerment tools such as the opt-in or opt-out setting that the government says it will require under this bill must be prominent, accessible, easy to use, functional, available on an ongoing basis and not affecting the functionality of the service or impairing the user experience. Why is this important? We know that tech companies will do what they can to exploit loopholes and keep users on their algorithm. After the EU required Meta to give users an option to switch off the algorithm, the company hid the setting, removed significant functions from the non-algorithmic feed and automatically switched users back to their algorithmic feed every time they closed and reopened the app. This malicious compliance is unacceptable. To ensure Australians have genuine control to switch their algorithm on and off, the setting must be prominent, accessible and easy to use and not affect the functioning of the app. Putting these standards in the primary legislation and not just in a ministerial determination strengthens these obligations and gives notice to social media companies that this is something they must comply with in good faith.

I also note that I had intended to move amendments to increase penalties for noncompliance with the digital duty of care. These amendments will now be moved by the member for Bradfield. The current penalties that apply to companies that do not comply with the digital duty of care are roughly $100 million. This would be about one-twentieth of one per cent of Meta's revenue last year—too small to change the behaviour of some of the wealthiest companies in the world. The EU, UK, Canada and New Zealand have either implemented or proposed fines that range from three to 10 per cent of global revenue. We should do the same. A big stick like this is essential to get these companies to change their behaviour. I would also like to see a private right of action added so that Australian users, not just the regulator, could enforce this new duty.

I would like to thank thousands of Curtin constituents who informed my work on this bill by responding to my community survey, the UWA students who attended my workshop on this issue and the many organisations across the technology ecosystem who engaged with me and my office. I commend the government for its work on this bill, and I thank the minister, who has indicated that the government will support these amendments. I commend these amendments to the House.

Zali Steggall

I rise in support of the amendments moved by the member for Curtin and agree wholeheartedly on the importance of ensuring these attachment bots be included in relation to online harms. But I also have a question for the minister because, whilst there's a strong focus on protecting under 18s from harms online, there are other foreseeable harms that we're seeing occur and are not being addressed in this digital duty of care. My question for the minister in this consideration in detail process is: why has the government failed to explicitly include harmful AI generated deepfakes in its duty of care legislation? If the purpose of this legislation is to prevent foreseeable online harm, why are we failing to address one of the fastest-growing threats?

We've got an opportunity to get ahead of an emerging threat, yet the government is leaving a significant gap in online safety protections. Artificial intelligence can now make someone appear to say something they never said, do something they never did and participate in an event that never happened. We've already seen how this technology is being misused in Australia. Deepfake video circulating, for example, of Senator Jacqui Lambie was falsely depicting her criticising another party's voters. That results in threats from those voters back to her. At a time when we're talking about a heightened security environment for members of parliaments and the public at large, we're allowing this to continue to be legal.

The amendment I had wished to propose to this effect sought to address deepfakes depicting identifiable people without their consent where the content is capable of misleading viewers and likely to cause serious harm. We know the government has looked at this from the nudify apps perspective, but the harm goes beyond that. Harm to a person, whether they are under 18 or over 18, where their image, likeness and voice are used by AI technology to generate deepfake content that is then pushed through social media platforms and digital platforms and where there is no consent from that person that is readily identifiable should also be treated as a harm that these platforms should have to prevent and be aware of. So I ask the government to look at protecting all Australians, not just under-18s, from this use without their consent of their image, likeness and voice in relation to the creation of AI deepfakes. It is not just in the political space that this matters. This could ruin someone's business reputation. If someone is in the public eye in sport—we've seen Sportsbet do it with some actors as well. It is really important that the government address this gap.

I would also like to note my deep frustration with the process. This was an amendment that I had sought to get drafted, but, through lack of resources through the system, it has not been possible for the parliamentary drafting services to provide the draft of that amendment despite being asked to do this over a week ago. I find it deeply unacceptable for the parliament to not be able to provide adequate services to members.

Anika Wells

Let me first respond to the member for Warringah's questions. I can confirm that AI generated images that cause serious harm would be defined as adult cyberabuse. They can be taken down by the eSafety Commissioner. That would include deepfake nude images. The duty would also require platforms to have systems in place to prevent the harm from occurring.

The member is correct in that material and content that does not cause serious harm is not the subject of this bill. We feel that the Joint Select Committee on Artificial Intelligence that is currently on foot is meant to be looking into these particular issues, and we intend to incorporate those informed findings into the forthcoming Australian AI standards, which we believe is the more appropriate place for them. Other feedback is noted. I also thank the member for Curtin for her amendments, which the government will support.

On user empowerment tools, the government recognises the importance of making sure that the primary bill mandates that the user empowerment tools are accessible, easy to use, prominently available on the service, and functional and available on an ongoing basis. We support that. On attachment hacking, we also recognise that the use of AI chatbots and companions are creating risks to children's long-term development and it's horrifying. As sociologist Sherry Turkle observed: 'Social media came for our attention. Chatbots are coming for our attachment.' I've spoken in this chamber before about the boy who was found to have 50 different tabs open seeking companionship, and whilst some of those chat—

Colin Boyce

The time for debate has expired.

Milton Dick

The question is the amendments circulated by the honourable member for Curtin be agreed to.