All Women or No AI

In 2024, a viral social media trend asked women a simple question: would you rather encounter a bear or a man alone in the woods? Most chose the bear. The threat felt more honest.

That instinct is data too.

I spent 15 years in big tech. I left to found the Ethical AI Alliance because I kept watching the same thing happen: harm lands on women first, and the people who designed the system are never in the room when it does.

Two years ago I posted a short video saying “AI was built by men for men.” One million views. Thousands of comments. Women sharing reason after reason why they agreed. Then came the misogynistic replies, basically proving the point.

Strip away the tech, and the top three issues women face globally are brutally consistent: gender-based violence, economic inequality, and health disparities. One in 3 women experiences physical or sexual violence. Women earn 77 cents for every dollar men make. Women are more likely to be dismissed, misdiagnosed, or simply not believed.

If AI were built for women, a significant part of it would be addressing these. So where is it? Follow the money.

AI firms captured $258 billion in venture capital in 2025 alone. Roughly 1% goes to AI for Social Good. And when you look specifically at what women actually need most, gender-based violence tools, economic inclusion, women’s health, the numbers become almost impossible to defend. My own bottom-up estimate puts total AI funding directed at these three areas at somewhere between $200 million and $1.5 billion globally. Even at the high end, that is less than 1% of total AI investment going toward the issues that are most likely to get women killed, keep them poor, or get them ignored by a doctor.

This matters for SDG 5 directly. But it also matters for every other goal. You cannot achieve SDG 3 when women’s health is an afterthought in AI development. You cannot achieve SDG 8 when economic tools are built for someone else’s economy. Gender is not a standalone goal. It runs through all of them.

Less than 1%.

But the funding gap is only part of the picture. The deeper problem is design.

Between July 2025 and January 2026, Grok generated hundreds of thousands of sexualized images per day. Three million in under two weeks. 81% of them women. The public debate asked why the filters failed. At the Ethical AI Alliance, we think it is the wrong question. Grok didn’t malfunction. It was built with zero-cost generation, permissiveness as a product feature, and distribution infrastructure reaching 650 million users. When outcry came, xAI restricted access to paying subscribers. The response to mass sexual abuse imagery was to monetize it.

This is what a design failure looks like. A system built without accountability to the people it would harm, operating exactly as designed. xAI built the model. X hosted the infrastructure. Individual accounts did the generating. And when the harm scaled, each pointed at the others. That fragmentation is not an accident; it is the architecture.

And yet the imagination exists. The builders exist. María Salguero, a geophysicist in Mexico, built an open-source femicide map, meticulous, painstaking, a lifeline for families and activists, largely alone and underfunded. Safecity mapped street harassment through crowdsourced reports. Callisto helped survivors of sexual assault identify repeat offenders. These tools worked. They struggled because the money never followed, because the people making funding decisions were not the people these tools were built to protect.

The solutions exist. The will to resource them does not.

What needs to change is where we intervene. Governance that only responds to harm after it scales is not governance. We are only cleaning up. The intervention point is upstream: before the system ships, not after the headlines. Black and Brown women are misidentified by facial recognition. Indigenous women face data erasure. Women in conflict zones live under AI-assisted surveillance. In each case, accountability was never built in. In each case, the harm was predictable.

At the Ethical AI Alliance, we work from a simple premise. If the technology does not start from the women most harmed, it will not reach the rest.

The women building the right tools are already here. María is still updating her map. The survivors who built Callisto were right. The developers behind SafeNess had the idea years before anyone with money paid attention. The technology works. What is needed now is funders and builders with the courage to go upstream, before the harm, before the cleanup.

That is the dare on the table.

All women or no AI.

Author: Asma Derja, Founder & Director, Ethical AI Alliance

References

• WHO (2021). Violence against women prevalence estimates. World Health Organization.
• UN Women (2023). Facts and figures: Economic empowerment. United Nations.
• OECD (2026). AI firms capture 61% of global venture capital in 2025. Organisation for Economic Co-operation and Development.
• World Economic Forum. AI for Social Good investment estimates.
• Internet Watch Foundation (2026). Reporting on Grok-generated child sexual abuse material.
• Safecity. Crowdmapping platform for sexual harassment. safecity.in
• Callisto. Survivor-led platform for sexual assault reporting. projectcallisto.org
• María Salguero. Femicide map of Mexico. feminicidiosmx.com
• Derja, A. and Albakri, S. (2026). Designed to Harm: Grok, Sexual Abuse Imagery, and the Architecture of Technopatriarchy. Ethical AI Alliance. ethicalaiallian`ce.org/opinion/Blog%20Post%20Title%20One-3zaa9-zlxng-n5tpp