When the algorithm answers, who does it answer for?

Kate Crawford, in Atlas of AI, calls these systems registers of power: not just tools that reflect inequality, but infrastructures that produce it, through the rare-earth mining that powers their hardware, the underpaid annotation labour that trains them, the data extracted without consent, and the institutional decisions they then automate. Abeba Birhane, an Ethiopian cognitive scientist whose audits of LAION and other foundation datasets have rewritten the field, has shown that the datasets behind some of the most celebrated generative models contain images and captions of staggering racism, misogyny, and harm. We are training the future on the worst of the past, unless we deliberately choose otherwise.

II. Bias does not arrive as an error. It arrives as a tone.

The most consequential finding in the UNESCO research is not the magnitude of any single output but the principle of repetition. Generative AI does not need to be wrong dramatically; it only needs to be wrong consistently. When language models repeatedly cast women in domestic settings, men in executive ones, certain accents as suspect, certain names as risk, the bias does not arrive as an error to be flagged. It arrives as a default. It arrives, more precisely, as a tone: the ambient register against which all subsequent claims are measured.

This matters acutely in mission-driven work because language is the medium through which the work travels. Risk assessments become evidence. Case notes become eligibility decisions. Grant applications become funded programs. Communications become the institutional voice that reaches communities. If the instruments generating that language are quietly tilted, the institutions built atop them tilt with them, invisibly, but cumulatively. Consider a practitioner drafting a risk assessment with AI assistance: the system that nudges, almost imperceptibly, toward describing one mother as volatile when it would have called another stressed. The proposal that, when generated, defaults to passive constructions about migrant women: they were brought, they were assisted, they were resettled, rather than constructions that grant agency. Las palabras que nos eligen también nos forman. The words that choose us also shape us.

The mechanism here is older than artificial intelligence. Catherine D'Ignazio and Lauren Klein, in Data Feminism, observe that "data is never raw". It is always cooked by someone, in some kitchen, for some meal. The proposition extends, with devastating accuracy, to generated text. Someone wrote the recipe. The question worth asking, before we adopt any new tool in a mission-driven setting, is whether they were cooking for the people we serve, or for an audience we have never met.

III. The shape of the gap

The conversation about artificial intelligence is global, but its harms are not evenly distributed across geographies, economies, or genders. The gap is increasingly legible in the data. Good Things Australia, in its 2025 national survey, reported that one in three Australian women struggle to distinguish AI-generated content from human-made content. The number rises to 39% among mothers, 49% among grandmothers, 40% among women with disability, and 35% among culturally and linguistically diverse women. Deloitte's 2025 TMT report, cited by the Australian Marketing Institute, found that 50% of Australian women were using generative AI at work, compared with 70% of men. The disparity is not merely one of access. It is a disparity of confidence, trust, and time.

The Australian numbers mirror international patterns. Whether the lens is global, national, or local, the populations most likely to be misread by these systems are also the ones most likely to be excluded from the rooms where the systems are designed.

Local feminist institutions are beginning to articulate what this means for governance. In March 2026, the Victorian Women's Trust launched its inaugural Feminist Researcher in Residence program with artificial intelligence as its defining inquiry, naming, in its founding statement, that AI systems are too often designed in rooms where the needs of women, girls, and gender-diverse people are absent, and that without a seat at the table, the community risks being left out of the equation entirely. The framing matters far beyond Victoria. It treats AI not as a technical question to be delegated to vendors, but as a structural one: a question of whose authority shapes the architecture before the architecture shapes everyone else. The next great feminist fight, on the evidence, will be about data and the systems built from it. Whoever stays out of that conversation will be governed by it.

The argument is not that artificial intelligence should be excluded from mission-driven work. The argument is that it cannot be deployed as if its harms were evenly distributed. They are not. They concentrate, with mathematical reliability, on the populations our missions already strive to keep safe. This is precisely why human-in-the-loop oversight, organisational accountability for AI-generated content, and proper training for the people working with these tools are not optional features. They are the conditions that make adoption defensible in the first place.

IV. Designing for the work AI cannot do

Productivity decks have a habit of describing AI adoption as a subtraction: fewer hours spent on email, fewer hours spent on summaries, fewer hours spent on first drafts. The arithmetic is incomplete. AI tools, however good, cannot read the room. They cannot hold a story across a long arc of trust. They cannot translate a culturally specific silence into a culturally appropriate response. They cannot tell when a sentence that scans correctly on the page would land wrong on the ear of the person it describes. That work is human, and it is the work that distinguishes mission-driven practice from any other kind.

A century of feminist political economy, from Silvia Federici's writings on reproductive labour to Arlie Hochschild's second shift to Tressie McMillan Cottom's recent work on the political economy of inclusion, has documented a pattern worth keeping in mind here: when new tools are introduced into systems of care, the relational and corrective work that holds the system together can quietly become invisible. The pattern is not anyone's fault. It is structural, and it is what good design and good leadership are for. Organisations that introduce AI thoughtfully account for that human work in workloads, in role descriptions, in budgets, and in their definition of what success looks like. They treat the time required for review, contextualisation, and judgement as core to the work, rather than as friction to be optimised away.

Sasha Costanza-Chock, in Design Justice, formulates a question that can usefully sit on every procurement table: "Whose labour is the system designed to save, and whose labour is it designed to extract?" Holding that question alongside questions of cost, capability, and scale produces better decisions, and better tools.

Whose labour is the system designed to save, and whose labour is it designed to extract? Sasha Costanza-Chock, Design Justice (2020)

V. Five questions for the procurement table

The single most useful contribution a feminist analysis makes to AI adoption is a revised set of questions for the procurement table. Standard procurement frameworks ask whether a tool works, whether it scales, whether it saves money. A feminist framework adds another register, and refuses to skip it.

1. What problem is this tool solving, and for whom?

A clear answer to this question is the foundation of every other answer that follows. If the underlying problem is one of capacity or staffing, the question beneath the question is whether automation alone is the appropriate intervention, or whether automation works best alongside investment in people.

2. Whose knowledge shaped this system, and whose is missing?

Foundation models are trained predominantly on English-language internet text, a corpus whose composition is well-documented and whose absences are not random. The communities most absent from training data are often the same communities mission-driven organisations exist to serve. Knowing this is not a reason to walk away. It is a reason to design carefully, validate thoroughly, and supplement what the tool does not know.

3. Who carries the risk if the model is wrong?

In a marketing department, a hallucinated output is an embarrassment. In a system that touches a person's safety, housing, or care, it is a matter of trust and consequence. The threshold of acceptable error must be calibrated accordingly. A model that is good enough for a technology company's internal workflows is not, by transitive logic, automatically good enough for ours.

4. Does this support frontline judgement, or replace it?

Tools that grant practitioners better information, more time, and more authority to bring their own expertise to a decision are categorically different from tools that route decisions around them. Both are sold under the same heading. The first kind is what good adoption looks like. The procurement question is not is this AI? but which kind?

5. Who benefits from this tool, and who might be left out?

This is the question that most often goes unasked, and the one that most determines whether an adoption serves a mission or quietly drifts away from it. Naming the beneficiary clearly is the first step. Naming who might be left out, or read incorrectly, or under-served by the same tool, is the second. A tool that demonstrably extends a mission's reach to people who have historically been underserved is one of the most powerful instruments a sector can deploy. A tool that quietly contracts that reach is the opposite, even if its dashboards say otherwise.

VI. The case for responsible adoption

It is essential to say plainly: not every concern about artificial intelligence is a reason to refuse it. The benefits of AI, used responsibly, are substantial and worth pursuing. Tools that translate across languages in real time can extend services to communities long underserved by monolingual systems. Tools that summarise large bodies of literature can give small organisations access to evidence that was previously the privilege of well-resourced institutions. Tools that automate administrative burden can free practitioners to spend more time with the people they serve, not less. Tools that detect patterns across data at scale can surface inequities that would otherwise remain invisible. The promise is real, and missing it would itself be a failure of mission.

What a feminist framework offers is not a refusal of these tools, but a discipline for adopting them well. The discipline is simple to state, and harder to practise: be clear about who benefits from each adoption decision, and be equally clear about who might be left out, misread, or under-served. Resource the human work of review, judgement, and accountability properly. Train the people using the tools. Document what the tools cannot do. Stay close to the communities the work is for. None of this slows progress. It is what progress, properly understood, requires.

VII. What feminists are already building

It would be intellectually dishonest to mount any account of the feminist conversation about AI without acknowledging the construction work already underway. The feminist AI field is not a complaint department. It is a building site, and the structures going up in it are worth examining seriously.

Chayn, founded by Hera Hussain, develops survivor-centred digital infrastructure across 195 countries, including assisted letter-writing systems for victims of intimate-image abuse seeking the removal of non-consensual content. The A+ Alliance, led jointly by Women at the Table and Code for Africa, has incubated SafeHer; SOF+IA, an Argentinian system for flagging gender-based political violence online; and AymurAI, which structures judicial data on gender violence cases for use by court administrators. Caroline Sinders's Feminist Data Set is a multi-year participatory project examining what it means to construct a dataset by consent rather than by extraction. Pollicy's Afro-Feminist AI principles, developed across the African continent, articulate the operational difference between symbolic inclusion and material participation.

None of these projects claims to have resolved the contradictions of the field. All are working through the same governance, labour, and consent questions the wider sector now faces. What they share is a refusal to treat artificial intelligence as a finished product to be received from a vendor, and a sustained commitment to building tools that remain accountable to the people they affect. The shift is from AI as product to AI as accountable infrastructure.

VIII. A practical agenda, for the people in the meeting

This essay is reflective, not prescriptive, but readers will reasonably expect to leave with operational instruments. What follows is a short discipline: five commitments offered in the order an organisation might attempt them. They are insufficient on their own. They are the beginning of a longer practice.

One. Begin every AI conversation with mission, not tools. If a use case cannot be defended in the language of organisational purpose without resorting to efficiency as a proxy virtue, it deserves more thought before it deserves a contract.

Two. Map benefit and risk by community, not by department. The relevant question is not which team uses this tool but which people does it touch, who among them benefits most, and who might be most likely to be misread by it? Both halves matter.

Three. Treat human review as a design requirement, not a compliance gesture. A practitioner with the time and authority to bring their judgement to a tool is what makes the tool trustworthy. Resourcing that time is part of resourcing the work.

Four. Train the people using the tools. The single most consistent finding across responsible AI research is that organisations that invest in literacy and judgement at all levels of the workforce make better decisions, faster, and recover from mistakes more cleanly than those that do not.

Five. Include affected communities in governance, not in feedback forms. There is a structural difference between consulting people about a system and granting them authority over it. The first is a public relations exercise. The second is a feminist commitment.

IX. Closing, on speed and what it earns

It is necessary to name, in closing, a conviction that haunts every conversation about artificial intelligence in mission-driven work. Speed is not, by itself, progress. Adoption is not, by itself, transformation. Efficiency, considered alone, is a measurement, and the political question is always the same: efficient at what, for whom, at whose expense, and over what time horizon?

The reason the social, education, health, and care sectors exist is that certain forms of human need cannot be automated, scaled, or optimised without becoming something other than what they were. Care is one of these forms. Trust is another. Witness is a third. The relationships that sustain this work, between practitioners and communities, between institutions and the people they have historically failed, between workers and the cultural literacies they accumulate over decades, are not in the training data. They cannot be retrieved by prompt. They are built, slowly, and they are the asset most worth protecting in any adoption decision.

The feminist tradition offers something more useful, in this moment, than warning. It offers a method: a refusal of the false binary between innovation and values; a discipline of asking who benefits, who bears the cost, whose voice is heard, and whose voice carries when an automated system speaks with institutional authority. The method is older than computing. It will outlast the present technological enthusiasm. It applies, with surgical precision, to the question now in front of every leader making decisions about these tools.

The trajectory of mission-driven work over the next decade will be determined less by what the technology becomes than by what its institutions choose to govern. Artificial intelligence will not become equitable through good intention alone. It will become equitable, where it does, through the unromantic, patient work of building accountability into the architecture from the start. That work is a feminist undertaking, whether or not it is named one. And it is, quietly, already underway.

There is a phrase that travels across feminist movements in different forms and different languages: lo que se construye sin nosotras, se construye contra nosotras. What is built without us is built against us. The sentence predates artificial intelligence by a century. It will outlast the present iteration. The remaining question, and it is genuinely open, is whether we choose to be in the room while the building is happening.