August 2026

Executive Summary
‍Artificial intelligence is reshaping how people communicate, find information, form relationships and participate in digital life. It is also changing the landscape of sexual violence. Generative technologies are facilitating AI-generated child sexual abuse material, sexual deepfakes and other forms of AI-enabled image-based abuse. Conversational AI and AI companion applications are also becoming part of the digital environments in which people explore relationships, sexuality, consent, gender and respect.
‍
Policy responses have largely focused on regulation, online safety, criminal misuse, privacy and ethical governance. These remain essential. However, regulation alone addresses only part of the challenge. Primary prevention also requires attention to the social, cultural and technological conditions that influence attitudes, beliefs and behaviours before violence occurs.

This commentary advances a cautious proposition: AI-mediated digital environments should be recognised as an emerging area of policy significance for the primary prevention of sexual violence. Artificial intelligence does not create the underlying drivers of sexual violence, and its influence should not be overstated. However, as conversational systems increasingly participate in everyday exchanges, they may reinforce, leave unchallenged or help disrupt ideas about gender, entitlement, objectification, consent and respectful relationships. This warrants research and deliberate policy attention while these technologies and their social uses are still evolving.

The current evidence does not establish that conversational AI can prevent sexual violence. Evidence is strongest in adjacent areas: detecting and disrupting child sexual abuse material, interrupting some pathways towards offending and reducing the continued circulation of intimate images. These activities span early intervention, response and harm reduction; they should not be misrepresented as primary prevention. Evidence about the longer-term influence of conversational AI on violence-supportive attitudes, behaviours and social norms remains limited.

The role of AI in relation to victim-survivors also requires clear boundaries. People may use conversational AI to understand whether an experience may constitute sexual harm, locate reliable information or identify available services. Systems should therefore be designed to provide accurate, current information and safe pathways to human services. Any such role must remain tightly bounded. AI should not provide counselling, crisis response, therapeutic support, risk assessment or legal advice, or replace specialist services and trusted human relationships.

The appropriate response is neither technological optimism nor technological resistance, but stewardship. Specialist sexual violence expertise should inform how relevant technologies are designed, governed, tested and evaluated. Governments and technology developers should also resource meaningful participation by victim-survivors, specialist services, prevention researchers and affected communities rather than shifting responsibility for technological safety onto those contributors.

Priority actions include embedding sexual violence prevention expertise in the development of AI policy, standards and safeguards; establishing a multidisciplinary research agenda; requiring developers to test for foreseeable sexual violence and child exploitation risks; strengthening access to reliable information and human services; and evaluating outcomes in terms of safety, agency, equity, social norms and unintended harm, not technical performance alone.

Artificial intelligence will continue to evolve. The question is whether our approach to preventing sexual violence will evolve with it.

Purpose and Scope
‍Primary prevention has continually evolved as understanding of sexual violence, its drivers and the environments in which it develops has deepened. Prevention now reaches beyond individual awareness to engage with the settings in which social norms and expectations are created and reinforced, including schools, workplaces, universities, sporting organisations, communities and digital platforms.

Artificial intelligence introduces new capabilities into digital environments. Unlike technologies that primarily retrieve or display information, conversational AI can respond through natural-language dialogue and permit follow-up questions and private exploration of sensitive subjects. Its fluent responses may be perceived as authoritative even when they are inaccurate. This changes how information is encountered and raises new questions about how beliefs and social norms may be reinforced or challenged.

This commentary does not argue that AI is a distinct prevention setting equivalent to a school, workplace or community. It understands conversational AI as an increasingly influential feature of broader digital environments, one that can mediate access to information and participate in exchanges about relationships, sexuality, consent, gender and wellbeing.

Artificial intelligence is also a broad term. This paper distinguishes between:
• generative AI, which creates text, images, audio or video;
• conversational AI, which interacts with users through natural-language dialogue;
• AI-assisted detection, which classifies or identifies potentially harmful material or behaviour;
• recommendation systems, which rank and promote digital content but are not necessarily generative AI; and
• other automated digital tools, including image-hashing and matching systems that may not use AI.

These distinctions matter because the technologies perform different functions and carry different risks. They also sit at different points across the primary prevention, early intervention, response and recovery continuum.

Given the pace of technological change and the limited evidence regarding longer-term social effects, this paper is a policy contribution rather than a definitive position. Its purpose is to identify why AI-mediated environments warrant attention from prevention policy, clarify the limits of current evidence and propose principles for responsible engagement.

Artificial Intelligence and the Changing Landscape of Sexual Violence
‍Artificial intelligence does not create the underlying social drivers of sexual violence. Sexual violence is shaped by interacting factors including gender inequality, power imbalances, harmful social norms, rigid gender roles, entitlement, objectification and broader structural inequities. Technology operates within these conditions; it does not sit outside them.
‍
AI is, however, changing how sexual violence can be perpetrated, facilitated, amplified and experienced. Generative systems can produce realistic sexual images, audio and video with limited technical expertise. These capabilities are being used to create AI-generated child sexual abuse material, sexual deepfakes and other forms of image-based sexual abuse. They can increase the speed, scale, accessibility and reach of abusive behaviour, while complicating detection, investigation and removal (eSafety Commissioner, 2023; Internet Watch Foundation, 2026; UNICEF, 2026). Internet Watch Foundation operational data provide one indication of this escalation: its analysts assessed 8,029 AI-generated images and videos as depicting realistic child sexual abuse in 2025. This figure describes material encountered and assessed by that organisation; it is not a population-prevalence estimate (Internet Watch Foundation, 2026).

This language is important. A sexual image created or altered using AI can cause real harm even when the depicted act did not occur. Where a child is depicted, the material should be recognised as child sexual abuse material, not minimised as merely synthetic or fake. Where an identifiable person is depicted without consent, the conduct should be understood within the framework of image-based sexual abuse (eSafety Commissioner, 2026b; UNICEF, 2026).

AI-mediated environments may shape people’s exposure to content relevant to harmful social norms, although findings should not be generalised across all systems. In a 2024 algorithmic-modelling study of TikTok, the proportion of recommended videos containing misogynistic content increased from 13 per cent to 56 per cent over five days for the modelled accounts (Regehr et al., 2024). The study found increased exposure to misogynistic content within a particular platform and research design; it does not establish how that exposure affects attitudes or behaviour, or whether the findings apply to recommendation systems more broadly.

Australian evidence also confirms that AI assistants and companions are already part of some children's digital lives. In an eSafety survey of 1,950 Australian children aged 10 to 17, 78 per cent had used an AI assistant and 8 per cent had used an AI companion. Among children who had used either type of tool, 54 per cent reported at least one personal or social use, 20 per cent reported a potentially inappropriate or harmful interaction and 32 per cent reported sharing personal or potentially sensitive information (eSafety Commissioner, 2026a). These findings do not imply that children should use these tools. Rather, they document existing use and reported experiences; they do not establish how that use affects sexual-violence-related attitudes or behaviour. Separate eSafety guidance identifies potential risks associated with AI companions, including sexualised interactions, unhealthy relationship expectations, privacy risks and emotional dependency (eSafety Commissioner, 2025).

These technologies do not determine behaviour. Their influence is shaped by system design, training data, commercial incentives, safeguards, user behaviour and wider social conditions. The relevant prevention question is therefore not whether AI causes sexual violence. It is whether AI is becoming sufficiently influential within digital life that the beliefs and norms it reflects, reinforces or leaves unchallenged now warrant deliberate attention.

Why AI-Mediated Environments Matter to Primary Prevention
In this commentary, AI-mediated environments are digital spaces and platforms in which AI systems influence the content people encounter, the interactions they have and the experiences available to them. Contemporary prevention science recognises that individual attitudes and behaviours develop within relationships, institutions, communities and wider social systems. The World Health Organization's settings approach similarly recognises that health and wellbeing are created within the environments in which people live, learn, work and interact (World Health Organization, 1986). In Australia, Change the Story identifies the need to transform the social norms, practices and structures that reinforce gender inequality and violence against women (Our Watch, 2021).

Digital environments are already part of this prevention landscape. They influence how people learn about relationships, interpret social expectations and encounter ideas about gender, sexuality and consent. Conversational AI adds a new dimension because it does not simply present information: it participates in an exchange.

This interaction warrants particular attention where users ask sensitive questions or seek personal or social advice. Australian survey data show that some children already use AI assistants and companions for these purposes, but do not establish how such interactions affect attitudes or behaviour (eSafety Commissioner, 2026a). A system's response could challenge a harmful assumption, provide neutral information, validate an unhealthy belief or fail to recognise coercion and abuse. These are foreseeable possibilities to be tested, not established prevention effects.

There is not yet sufficient evidence to determine the longer-term effects of these interactions on attitudes or behaviour. Nor should conversational AI be treated as a substitute for comprehensive respectful relationships education or sustained, whole-of-setting prevention. However, waiting for definitive evidence before examining foreseeable risks may allow harmful design features and patterns of use to become entrenched.

The experience of social media provides a cautionary analogy for prevention lag: technologies can become deeply embedded in relationships and culture before their wider social effects are adequately understood. AI presents an opportunity to ask prevention-informed questions earlier, without assuming either harm or benefit in advance of the evidence.

Recognising AI-mediated environments as relevant to prevention therefore does not mean deploying AI as a prevention program. It means asking prevention-informed questions during policy development, design and evaluation:
• Does the system reproduce gender stereotypes, entitlement, objectification or victim-blaming?
• How does it respond to questions involving consent, coercion, age, power and respectful relationships?
• Does it validate hostility or direct users towards increasingly harmful material?
• Are safeguards effective for children and people who may be vulnerable to manipulation?
• Can users distinguish general information from professional, therapeutic or legal advice?
• Are affected communities involved in deciding what safety and accountability should require?

These questions extend established prevention principles into changing digital environments without assuming that technology itself can prevent sexual violence.

Helping victim-survivors find reliable information
‍Conversational AI can be used to seek sexual-health information and personal or social advice, but the evidence reviewed for this paper does not establish how often victim-survivors use it specifically to interpret sexual harm or locate services (Döring et al., 2025; eSafety Commissioner, 2026a). The relevant policy question is therefore precautionary: if a person asks a system about an experience that may involve sexual harm, how can the risk of misleading or unsafe information be reduced and pathways to reliable public information and human assistance strengthened?

Where a system encounters questions that may relate to sexual harm, its appropriate role should be limited and clear: helping a person locate reliable public information, identify emergency options and find relevant specialist services. Information should be current, jurisdictionally appropriate and transparent about its limits.

Conversational AI should not be positioned as providing counselling, crisis intervention, therapeutic support, individual risk assessment or legal advice. It should not encourage people to disclose unnecessary personal information, make decisions on their behalf or become a gatekeeper to human services. People must retain choice over whether to use the technology, and non-digital and human pathways must remain available.
‍
This distinction is central: the opportunity is to make trustworthy information and pathways easier to find—not to automate specialist support or replace the relationships upon which safety, recovery and accountability depend.

What the Current Evidence Shows
‍Evidence concerning AI and sexual violence is developing but remains uneven. The most established applications sit within detection and disruption, not primary prevention.

Detection and disruption
‍AI-assisted tools can help identify, classify and investigate child sexual abuse material and analyse patterns within online communications. A 2025 Australian Institute of Criminology rapid evidence assessment identified 33 empirical studies at the intersection of AI and child sexual abuse. All eligible studies concerned the use of AI for prevention and disruption, particularly detecting or investigating child sexual abuse material and suspected offenders. None examined the use of AI in child sexual abuse offending, highlighting how incomplete the evidence base remains (Wolbers, Cubitt, & Cahill, 2025).

These technologies may strengthen investigative capacity and help disrupt the continued distribution of abusive material. However, they also raise questions about privacy, bias, false positives, transparency and human oversight. Detection is a critical response to harm, but it should not be described as evidence that AI changes the underlying drivers of sexual violence.

Targeted early intervention
‍Technology can also be used to intervene when a person's online behaviour indicates a possible intention to seek child sexual abuse material. The reThink initiative, developed by the Internet Watch Foundation and the Lucy Faithfull Foundation and deployed on Pornhub UK, combined warning messages with a chatbot that directed users towards the Stop It Now! service. Its independent evaluation found a reduction in searches using potential child sexual abuse material-related terms and some engagement with further information and support (Scanlan et al., 2024).

reThink is best understood as a targeted early-intervention and deterrence initiative. It also demonstrates the potential of automated conversational tools, rather than establishing that generative or conversational AI prevents offending. Its lessons should be interpreted carefully and tested in other contexts.

Response and reduction of continuing harm
‍Services such as StopNCII.org and the National Center for Missing & Exploited Children's Take It Down enable people to create digital fingerprints, or hashes, of intimate images so participating platforms can identify and restrict matching content. These are important safety tools that can reduce continued distribution and help people regain some control over their digital identity.

They are not, however, examples of AI primary prevention. Their core processes use image hashing, and they operate after an image already exists. Their relevance lies in demonstrating that digital systems can be intentionally designed around privacy, agency and the reduction of continuing harm.

Primary prevention remains an evidence gap
‍The clearest potential contribution to primary prevention lies in the design and governance of AI-mediated environments themselves. Systems could be designed to avoid reinforcing violence-supportive attitudes, respond safely to abusive requests, provide accurate information about consent and respectful relationships, and reduce algorithmic amplification of misogynistic content. These are plausible directions, not established outcomes.

Research is needed to determine whether particular design choices influence knowledge, attitudes, behavioural intentions or social norms; for whom; under what circumstances; and with what unintended effects. Evaluation should extend beyond technical measures to consider safety, autonomy, trust, cultural relevance, equity and actual behavioural outcomes.

The measured conclusion is that AI already contributes to detection, disruption, targeted early intervention and harm reduction. Whether it can contribute meaningfully and safely to primary prevention remains an open question. That uncertainty is a reason for disciplined inquiry and early stewardship—not premature deployment or disengagement.

Stewardship in an AI-Mediated World
‍Stewardship recognises that specialist sexual violence expertise has value beyond responding to individual acts of harm. The sector understands coercion, grooming, trauma, disclosure, recovery, accountability and the social conditions associated with sexual violence. This expertise should inform technologies that may affect consent, relationships, gender equality, image-based abuse and child sexual exploitation. It should sit alongside established principles of human rights, transparency, accountability and responsible AI governance (OECD, 2024; UNESCO, 2021). The principles below are proposed policy safeguards derived from those frameworks and specialist-practice considerations; they are recommendations, not findings that have been demonstrated to prevent sexual violence.

Stewardship is a shared responsibility: governments retain responsibility for regulation and public policy; developers for product safety and accountability; researchers for independent evidence; and specialist services, victim-survivors and affected communities for contributing expertise that must be meaningfully heard and properly resourced.

This is consistent with the sector’s longstanding role in informing criminal justice reform, child-safe practice, respectful relationships education, workplace reform, prevention policy and responses to technology-facilitated abuse.

Specialist services are also likely to increasing encounter numbers of people affected by AI-generated image-based abuse and other forms of technology-facilitated sexual harm. The sector therefore requires sufficient capability to recognise these harms, respond appropriately and contribute confidently to relevant policy, research and technology-governance discussions.

Six principles should guide this work.
1. Human relationships remain central
‍Preventing and responding to sexual violence is fundamentally relational work. AI may assist with bounded administrative or information functions, but it should not replace specialist practitioners, educators, therapeutic relationships, community leadership or professional judgement.
2. Victim-survivor safety, dignity and agency guide decisions
‍Design and evaluation should be informed by victim-survivors through safe, paid and meaningful participation. Systems must protect privacy, minimise data collection, avoid retraumatisation and preserve choice. Personal disclosures and case material should not be used to train systems without a lawful, ethical and explicitly consented basis.
3. Information pathways are not specialist support
‍Technology may help people find accurate information and human services. It must communicate its limitations and should not present automated responses as counselling, crisis support, legal advice or individualised risk assessment. Human alternatives must remain visible and accessible.
4. Prevention and safety must be intentionally designed
‍Safeguards should be incorporated throughout product design, testing, deployment and monitoring. This includes examining how systems respond to harmful requests and to questions about consent, coercion, age and power, not relying solely on removing harmful content after it has been produced.
5. Evidence, independence and accountability are essential
‍Claims of safety or prevention should be independently evaluated. Findings, limitations and serious incidents should be reported transparently. Commercial novelty should not be mistaken for public value, and partnerships should protect the independence of specialist organisations and lived-experience contributors.
6. Human rights, cultural safety and inclusion are foundational
‍AI systems can reproduce existing inequities. Governance must account for the experiences of Aboriginal and Torres Strait Islander peoples, culturally and linguistically diverse communities, LGBTQIA+ people, people with disability, children and young people, and others whose experiences may be overlooked by dominant datasets and design assumptions. This requires self-determination, accessible participation and community-led approaches, not consultation after key decisions have already been made.

Priorities for Policy, Research and Practice
The immediate task is not to implement AI across sexual violence prevention and response. It is to establish the knowledge, governance and safeguards required to make sound decisions.

Embed prevention expertise within AI governance
‍Governments should incorporate sexual violence prevention, child safety and victim-survivor expertise into relevant AI policy, standards, impact assessments and public procurement. This should complement—not replace—strong regulation, platform accountability and criminal justice responses, and support commitments under the National Plan to End Violence against Women and Children 2022–2032 and the National Strategy to Prevent and Respond to Child Sexual Abuse 2021–2030 (Australian Government, 2021; Australian Government Department of Social Services, 2022).

Establish a multidisciplinary research agenda
‍Research should examine how conversational AI, AI companions and recommendation systems may reinforce or challenge attitudes relating to gender, consent, entitlement, objectification and victim-blaming. It should also test how people interpret AI-generated advice, which groups face particular risks and what safeguards are effective.

Longitudinal research will be particularly important to determine whether repeated interactions have any sustained influence on attitudes, behavioural intentions or social norms over time.

Require safety-by-design testing
Developers should test foreseeable risks relating to sexual deepfakes, child sexual exploitation, grooming, sexualised interactions, coercion and the validation of violence-supportive beliefs. Testing should include independent scrutiny and appropriately supported participation by affected communities.

Strengthen pathways to reliable information and human services
Where systems respond to questions about sexual harm, they should provide accurate, current and jurisdictionally appropriate information, communicate their limitations and offer clear routes to specialist human services. AI should never become the only or preferred gateway to assistance.

Resource sector and lived-experience participation
‍Specialist organisations and victim-survivors should not be expected to contribute expertise without appropriate resourcing, governance and support. Participation should be safe, paid, influential and transparent about how advice has shaped decisions.

Evaluate public value, not technological novelty
‍Evaluation should examine whether an initiative improves safety, access to reliable information, agency and equity; reduces opportunities for harm; or influences relevant attitudes and norms. It should also identify unintended effects, including privacy risks, exclusion, misinformation, over-reliance and displacement of human services.

‍Conclusion
‍Artificial intelligence is changing the ways sexual violence can be perpetrated, amplified and experienced. It is also becoming part of digital environments in which people seek information and explore relationships, sexuality, gender and consent.

The evidence does not currently support presenting conversational AI as a primary-prevention intervention. Current applications are strongest in detection, disruption, targeted early intervention and harm reduction. The longer-term influence of AI-mediated interactions on violence-supportive attitudes, behaviours and social norms remains uncertain.
That uncertainty should not result in either uncritical adoption or disengagement. It calls for stewardship: early, careful and properly resourced involvement by governments, researchers, technology developers, specialist services, victim-survivors and affected communities.

For victim-survivors, the boundary must remain clear. AI may help make trustworthy information and pathways to human assistance easier to find. It should not replace specialist support, professional judgement or human relationships.

Ultimately, this is not a call for artificial intelligence to replace human expertise. It is a call for human expertise to help shape artificial intelligence.

Evidence Note
‍This commentary draws on different forms of evidence that should not be treated as equivalent: peer-reviewed reviews and studies; Australian Government research and guidance; international policy and human-rights frameworks; program evaluations; and operational data from organisations that detect or respond to online abuse. Operational figures describe material encountered by the reporting organisation and do not establish population prevalence. Guidance documents identify recognised or foreseeable risks but do not, by themselves, demonstrate causal effects. Proposed stewardship principles and policy priorities are normative recommendations informed by the cited evidence and frameworks; they are not established primary-prevention interventions.