The harms of AI to women are usually discussed in the future tense and separately from the more dominant conversations about data centres, copyright, and productivity.
Meanwhile, abusive partners are using smart devices and remote car locking to control and surveil women. An estimated 98 per cent of deepfake videos are pornographic and 99 per cent of those feature women and girls. And Jobs and Skills Australia data shows 15 of the 20 occupations most exposed to automation are female-dominated.
These are just some of the findings in Gender and AI: Where’s the Harm?, a Research Snapshot released today by the Victorian Women’s Trust.
It identifies 20 harms that disproportionately affect women, girls and gender diverse people, and it’s described as the first known attempt in Australia to map these risks at a systems level.
The 20 harms range from the above to those that get far less attention, including reproductive surveillance, financial exclusion through algorithmic credit scoring, the ‘AI ethics mental load that’s failing women, homogenised beauty standards and young people’s growing reliance on AI to navigate relationships (full list below).
And far from being harms that might happen in the future, most of these harms are affecting women in Australia now.
Dr Elise Stephenson, Deputy Director of the Global Institute for Women’s Leadership at ANU and the Trust’s inaugural Feminist Researcher in Residence, led the research (and is pictured above).
I spoke to Elise for the Women’s Agenda Podcast, where we examined some of these risks, including around tech-facilitated abuse, AI companions and what they are teaching young people about relationships, the targeting of women in politics – and what it’s doing to democracy – and whether women should be using AI at all.
Elise’s starting point is that we may hear about deepfakes, job losses, and data centres as key issues separately, but we rarely discuss the full picture.
“What I failed to see was a linked-up picture, kind of a systemic look at what is happening when it comes to gender,” she told me.
The report says the harms cascade. When women are excluded from designing and governing AI (and currently, 88 per cent of leading AI researchers globally are men) the consequences multiply across safety, work and political representation.
“This isn’t speculative. It isn’t theoretical,” Elise said. “There are parts of the AI story that are a little bit theoretical, but there’s certainly a lot of harm that’s already happening.”
Take the car as an example. Being able to lock and unlock a vehicle from a distance is being reported as a means of control. Elise notes the same feature could help someone escape a violent situation, which is why blanket rules won’t work and why careful regulation, community standards, and education are needed.
We still don’t know how common some of the risks discussed in the report are, as prevalence data is missing.
Also missing is an extensive understanding of who is funding and coordinating some of the well-resourced AI-enabled campaigns and attacks against women in public life or women running for office.
Then there are the AI gendered harms that are harder to count. In the research, youth advocates have raised the issue of AI girlfriends and boyfriends, which are cheap, increasingly used and are ultimately built to comply – offering completely unrealistic ideas around what real relationships and conversations with people are all about.
As Elise asked, “What does it mean if you can kind of boss AI around and if AI doesn’t push back and maybe doesn’t set boundaries?”
Elise worries about how our interactions with AI subtly shift norms, because the full consequences won’t be visible for years. But that doesn’t mean we should wait for the full consequences before anticipating them and acting.
While this research snapshot has been released now, the recommendations aren’t due until early 2027 – and the release of these risks highlights the urgency of understanding the issue.
According to the Trust’s Executive Director, Dr Kirsten Abernethy, safeguards are scarce when it comes to AI, and Australia is relying on a “patchwork of anti-discrimination, privacy, and consumer protection laws that were never designed to address algorithmically-driven harm.”
“We are at a crossroads, and Australia is on the cusp of making policy decisions that will impact generations to come,” Abernethy said. “Without gender-responsive legislation, we are putting women’s safety, autonomy, political agency and economic security at risk.”
In my conversation with Elise, I asked about something I hear often: that a feminist response to AI is to refuse to use it at all. Elise believes women should use it, and that women carry more of the ethical worry about AI than the men in their lives, or than business. But AI learns from the people who use it, and when women withdraw, their voices drop out of the loop.
A feminist approach to AI, Elise says, starts by rejecting the idea that technology is neutral or that its path is inevitable. We put guardrails around nuclear technology and aviation, so why not AI?
“There’s a huge amount of things to be scared about and worried for when it comes to these gendered harms, but I think it’s also probably irresponsible for us to walk away,” Elise said.
But when it comes to AI leadership, I noted the lack of women in the room – something made stark during a press conference at the White House last week as AI leaders met to sign a “morally binding” document on safety. Not one woman was among the dozen or so leaders at the press conference with President Donald Trump, and only two women were among the 32 invited to the lunch where the issues were discussed.
Australia has an opportunity to address how it responds. The policy decisions that are being made now, and the people making them, should be asked where gender sits in their plans and who is in the room.
Gender and AI: Where’s the Harm? is available at the Trust’s website. You can hear the full conversation with Dr Elise Stephenson on the Women’s Agenda Podcast.
The 20 gendered harms of AI include:
Labour market inequality, disruption and job losses: female-dominated jobs disproportionately automated, and biased recruiting systems that suppress wages and widen pay gaps
Financial exclusion: inequality embedded in algorithmic credit scoring, and scam losses affecting women
Gendered digital exclusion: an AI literacy gap, unequal access to tools, and women’s inputs not training the models
Care burden intensification: women taking on the “AI ethics mental load” and more work monitoring AI outputs
Deepfakes and synthetic abuse: reputation damage, sextortion and sexual abuse
Technology-facilitated gender-based violence: doxxing, automated stalking and harassment, and coordinated abuse of women leaders
Surveillance and privacy violations: non-consensual surveillance, video and photos
Reproductive and bodily autonomy risks: surveillance of reproductive choices, predictive profiling of pregnancy, and misuse of health and fertility data
Gendered AI companions: normalising submissive AI personas and patterns of control, with offline impacts
Emotional and relational manipulation: AI reinforcing gender stereotypes and power dynamics in relationships
AI-assisted social dependence: reliance on AI for interpersonal communication, at a cost to social skills, empathy and trust
AI-mediated consumption: hyper-targeted marketing of harmful content, such as nudify apps targeted at men
Aesthetic standardisation: homogenised beauty standards through generated imagery
Civic and political exclusion: women withdrawing from public debate, and a chilling effect from online abuse
Synthetic social normalisation: AI-generated content overwhelming public discourse, with disinformation and echo chambers
Societal backsliding: discriminatory systems institutionalised across services, and AI content relied on in place of women’s real experiences
Under-representation and exclusion from AI: concentrated ownership, and women’s perspectives excluded from design
Biased AI systems: biased datasets, healthcare and legal inaccuracies, and algorithmic acceleration of the manosphere
Unchecked AI action: safety limits removed and loss of control of systems, with gendered patterns in who is making the decisions
Environmental and intergenerational harm: resource-intensive AI amplifying climate change, itself a multiplier of inequity