Submission 123
Professor Anastasia Powell
Submitter group: Individuals.
Position labels are not shown for individual submitters.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Argues that AI governance should treat gender-based violence, stalking and coercive control as foreseeable harms. Recommends safety by design, technology-neutral laws, platform detection and redress, survivor support, and recommender-system safeguards. Describes opportunities for detection and recovery, while requiring evaluated digital tools to complement specialist human services.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that AI governance should recognise gender-based violence and interpersonal abuse as foreseeable AI use cases.
I argue that AI governance should recognise gender-based violence and interpersonal abuse as foreseeable AI use cases.
Argues that AI does not create the drivers of abuse but can expand the means of perpetration and amplify its impacts.
AI does not create the underlying gendered drivers of abuse, but rather, it can expand the means through which abuse is perpetrated and amplify its impacts.
Argues that reform should be technologically neutral and harm-based, covering fabricated content across media and clarifying responsibility where AI agents facilitate harassment or stalking.
A technologically neutral, harm-based approach should address the creation, threat, use, showing and distribution of fabricated content across text, image, audio and video, and clarify responsibility where AI agents or third parties are deployed to facilitate harassment or stalking (Flynn et al., 2026).
Argues that AI also has potential to help platforms identify and disrupt abuse, and eSafety's industry guide identifies automated and AI-assisted detection as part of a layered safety response.
Importantly, AI also has potential to assist platforms to identify and disrupt abuse. eSafety’s Technology, Gendered Violence and Safety by Design industry guide already identifies automated and AI-assisted detection as part of a layered safety response
Argues that the AI Safety Institute should incorporate gender-based violence into its analysis and testing of emerging AI capabilities and risks.
The Australian AI Safety Institute should incorporate gender-based violence into its analysis and testing of emerging AI capabilities, risks and controls.
Argues that digital tools for survivor support should be independently evaluated and kept as an adjunct to, not a replacement for, specialist human support.
Government should support independent research and evaluation of AI-assisted technologies for safety planning, service navigation, psychoeducation and recovery, subject to trauma-informed co-design, privacy and data-minimisation requirements, transparent limitations, human escalation pathways and clear prohibitions on replacing crisis or specialist services.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Have the AI Safety Institute and relevant regulatory frameworks treat technology-facilitated sexual violence, family violence, stalking, coercive control and sexualised deepfake abuse as foreseeable AI harms, not peripheral misuse issues.
The Australian AI Safety Institute and relevant regulatory frameworks should treat technology-facilitated sexual violence, family violence, stalking, coercive control and sexualised Artwork ‘Luwaytini’ by Mark Cleaver, Palawa deepfake abuse as foreseeable AI harms, not peripheral “misuse” issues.
Require and support developers and platforms to assess, adversarially test and evaluate gender-responsive safety across foreseeable sexualised deepfakes, nudification, impersonation, sextortion, stalking and automated harassment.
Developers and platforms should be required and supported to undertake gender-responsive abuse risk assessments, adversarial testing and safety evaluations addressing foreseeable uses including sexualised deepfakes, nudification, impersonation, sextortion, stalking and automated harassment.
Continue to support and expand eSafety's work on safety by design.
Require Safety by Design across the AI lifecycle. Developers and platforms should be required and supported to undertake gender-responsive abuse risk assessments, adversarial testing and safety evaluations addressing foreseeable uses including sexualised deepfakes, nudification, impersonation, sextortion, stalking and automated harassment. eSafety’s leading and collaborative work in this area should continue to be supported and expanded.
Ensure technology-neutral, harm-based Commonwealth, State and Territory laws cover fabricated or altered image, video, audio and text content, its creation, threat, use and distribution, and AI-facilitated harassment, stalking or coordinated abuse.
Ensure laws are technologically neutral and harm-based. Commonwealth, State and Territory frameworks should adequately encompass digitally fabricated and altered content across image, video, audio and text, as well as its creation, threat, use and distribution; and the use of AI agents or other digital tools to facilitate harassment, stalking or coordinated abuse.
Require platforms to invest in AI-assisted detection of repeated and coordinated abuse, rapid survivor-centred reporting and removal, evidence preservation, transparent appeals and specialist human review.
Platforms should be required to invest in AI-assisted abuse detection alongside rapid, survivor-centred reporting and removal mechanisms, evidence preservation, transparent appeals and specialist human review. Detection should encompass patterns of repeated and coordinated abuse rather than only individual pieces of content.
Support independent research and evaluation of AI-assisted safety planning, service navigation, psychoeducation and recovery, subject to trauma-informed co-design, privacy, data minimisation, transparent limitations, human escalation and no replacement of crisis or specialist services.
Government should support independent research and evaluation of AI-assisted technologies for safety planning, service navigation, psychoeducation and recovery, subject to trauma-informed co-design, privacy and data-minimisation requirements, transparent limitations, human escalation pathways and clear prohibitions on replacing crisis or specialist services.
Resource specialist family and sexual violence services for brief-intervention and longer-term counselling, including complex technology-facilitated abuse cases, with appropriate digital-safety assessment, support and referral as needed.
Specialist sexual violence counselling and support sectors should be resourced to provide both brief-intervention counselling and longer-term counselling for more complex cases involving technology-facilitated abuse, alongside appropriate digital-safety assessment, support and referral as needed.
Establish formal mechanisms for the AI Safety Institute, eSafety and other regulators to consult victim-survivors, specialist family and sexual violence services and researchers alongside technical AI expertise.
The Australian AI Safety Institute, eSafety and other relevant regulators should have formal mechanisms for consultation with victim-survivors, specialist family and sexual violence services and researchers, alongside technical AI expertise.
Require developers to publish meaningful transparency data concerning AI-facilitated abuse and their systems' detection and response performance.
Developers should also be required to publish meaningful transparency data concerning AI-facilitated abuse and their systems' detection and response performance.
Address the role of recommender systems, generative AI and platform design in producing, amplifying and normalising misogynistic, sexually violent and other gender-inequitable content.
Australian AI and online safety policy should also seek to address the role of algorithmic ‘recommender systems’, generative AI and platform design in producing, amplifying and normalising misogynistic, sexually violent and other gender-inequitable content.
Require platforms to assess and mitigate amplification of misogynistic, sexually violent and gender-inequitable content, give users meaningful feed control, and strengthen safeguards for children and young people.
Australian AI and online safety policy should also seek to address the role of algorithmic ‘recommender systems’, generative AI and platform design in producing, amplifying and normalising misogynistic, sexually violent and other gender-inequitable content. Platforms should be required to assess and mitigate these risks, provide users with meaningful control over algorithmically curated feeds, and apply stronger safeguards for children and young people.
Build on existing primary prevention architecture, including Change the Story, to develop specific guidance for preventing gender-based violence in AI-enabled and digital environments.
Australia should also build on its existing evidence-based primary prevention architecture, including Change the Story, to develop specific guidance for preventing gender- based violence in AI-enabled and digital environments.
Prioritise survivor-led co-design and independent evaluation in government investment in AI-assisted survivor support.
Government investment should prioritise independent evaluation and survivor-led co-design rather than assuming that a tool is ‘trauma-informed’ because it uses supportive language.
Have the AI Safety Institute test gender-based violence risks and controls, potentially including lowered abuse barriers, safeguards and discriminatory detection risks, collaboratively with eSafety, specialist services, survivor-experts and researchers, beyond technical evaluation alone.
The Australian AI Safety Institute should incorporate gender-based violence into its analysis and testing of emerging AI capabilities, risks and controls. This could include testing whether models materially lower barriers to sexualised fabrication, impersonation, harassment, stalking or coordinated abuse; assessing safeguards against those uses; and examining whether safety technologies can reliably identify abuse without reproducing discrimination or creating new risks. This work should be developed collaboratively with eSafety, specialist family and sexual violence services, survivor-experts and researchers, and not solely through technical evaluation.
In AI regulation, constrain tools that make coercion, sexual exploitation, impersonation and harassment cheaper, easier or more scalable, require platform detection and disruption, and support evaluated applications expanding survivor choice, access and recovery.
Effective governance should therefore aim to constrain tools that make coercion, sexual exploitation, impersonation and harassment cheaper, easier or more scalable; require platforms to invest in the detection and disruption of foreseeable abuse; and support carefully evaluated AI applications that expand survivor choice, access and recovery.