In the First World. . .the oppressed, if given the chance . . .can speak and know their conditions. We must now confront the following question: On the other side of the international division of labor from socialized capital, inside and outside the circuit of the epistemic violence of imperialist law and education supplementing an earlier economic text, can the subaltern speak? – Gayatri Spivak, 1988

Artificial intelligence (AI)* is seemingly everywhere, as are its proselytizers. AI tools are not mere commodities; they are shapers of epistemologies. This essay elevates attention to the epistemic violence inherent in the aggressive promotion of new AI tools through both policy and proselytization.

This essay elevates attention to the epistemic violence inherent in the aggressive promotion of new AI tools through both policy and proselytization.

The massive and well-funded push toward AI adoption has numerous historic parallels: to cite just one, the push to hook the British working classes on tea and sugar during the Industrial Revolution. Then, as now, markets rely on cultural adaptation. Christian missionaries received support from nations with imperial ambitions as they played a role in erasing local knowledge and replacing it with an imperialist worldview on peoples and societies. As in those times, in today’s information era, economic hegemony relies on the promotion of reactionary norms regarding gender and social hierarchies. Voices of opposition must be silenced to complete the colonization of our knowledge ecosystems. Thus, the push for AI adoption is a particular threat to hard-won feminist epistemology, evidence, and knowledge.

In a previous essay, I described the failure of feminist responses to epistemic violence. This contribution proposes a way forward. We must simultaneously seek to contain a rapid and unregulated spread of AI use and to curate and preserve our own femisphere of intersectional feminist knowledge.

Missionaries at work

AI adoption is now big business, with a vast array of courses available to an anxious human workforce seeking to ‘adapt.’ Organizations, too, are targets for this new cottage industry; businesses, nonprofits, and philanthropies face a steady stream of “AI management” seminar options oriented entirely to ‘motivating’ teams toward AI adoption. Nevertheless, human anxiety is justifiably on the rise as examples pile up of human costs connected to AI adoption.

Rather than stop and correct the problems, tech companies are addressing public concerns by doubling down on proselytization. On 11 June, Anthropic announced the creation of Claude Corps, a program to embed its missionaries within nonprofit organizations. With imagery reminiscent of popular, government-supported public service programs such as the Peace Corps, the company is clumsily seeking to stand up an ersatz version of a public service ‘fellowship’ to carry out its commercial and ideological agenda.

With imagery reminiscent of popular, government-supported public service programs such as the Peace Corps, the company is clumsily seeking to stand up an ersatz version of a public service ‘fellowship’ to carry out its commercial and ideological agenda.

While there is a growing and thoughtful critique of the rapid and unregulated dissemination of a technology whose harms are too little understood, far too little of the commentary on AI even mentions the problem of epistemology. And to understand why we need a much more comprehensive focus on this issue, it’s important to understand the struggle for equity in epistemology in the first place. Scholars in post-colonial societies of the 20th century slowly and carefully worked to unearth and disseminate the histories of marginalized peoples in the face of decades or centuries of colonial erasure. These researchers in both core and peripheral countries included women and members of communities excluded on the basis of caste, class, race, and ethnicity.

The entrenched problem of whose knowledge had been buried was always gendered, as colonial patriarchy often layered onto existing hierarchies.

The entrenched problem of whose knowledge had been buried was always gendered, as colonial patriarchy often layered onto existing hierarchies. There are reasons why the world’s archetypal myths have required women to weave their stories, cook their wisdom into soups and potions, or, as in the tale of Cassandra, condemned their stories to forever be disbelieved and discounted. And this vast, subaltern knowledge is entirely invisible in today’s generative AI searches.

Encoding epistemic violence

This article intentionally starts with a quote from scholar Gayatri Spivak’s important article, Can the Subaltern Speak? Both history and epistemology were remaking themselves in the post-colonial world as scholars from the periphery, bit by bit, reclaimed the stories that their communities had kept hidden from the colonial gaze. It’s important for today’s digital sovereignty advocates to understand how Spivak’s article, like an avalanche, gathered force as it was disseminated, remaking the research landscape and elevating the impetus for the creation of new scholarly knowledge that embedded principles of equity and the representation of marginalized voices and communities.

As an overdue step, feminist scholars in social sciences used the opening to argue for the importance of simply collecting gender-disaggregated data. Decade by decade, large data sets that provide gender-disaggregated household data have been collected and, over time, have enabled important studies documenting gender gaps in food security, life expectancy, maternal mortality, educational attainment, and access to employment. Scholars focusing on marginalized and excluded populations have revealed similar gaps when it comes to race, class, ethnicity, disability, and other factors. These studies have, in turn, enabled smarter and more sensitive public policy investments.

Over time, the evidence has grown of the overall benefits for societies of ‘equitable’ public investment. And since returns on gender-equitable and socially inclusive investments have been conclusively proven, today’s reactionary politics require a complete remaking of what we know to be true in order to succeed. Concentrated control over AI search engines provides the opportunity to do exactly that.

And since returns on gender-equitable and socially inclusive investments have been conclusively proven, today’s reactionary politics require a complete remaking of what we know to be true in order to succeed.

Long-standing biases in knowledge management mean that intersectional feminist research, especially that produced outside the academy, remains disproportionately offline, scattered, or hard to search. As in every endeavor, feminist scholars start at a disadvantage due to historic inequities. Power and wealth dynamics within academic hierarchies have already shaped what evidence has been digitized. Curated spaces where such knowledge has been elevated, such as gender studies departments in universities, are in a precarious position due not only to chronic underfunding but also to new attacks from reactionary political forces. Generative artificial intelligence (AI) and algorithmic summaries are becoming ubiquitous and unavoidable on any search engine. Knowledge that is not digitized, optimized, and tagged for algorithmic amplification is simply impossible to find.

At the same time, harmful, deceptive, and misogynistic content is being amplified with AI hallucinations and false authority replacing the evidence produced and curated by and with marginalized voices and communities. Digital systems scrape data from an ecosystem where violent and sexualized images of Black and Brown women and children are set up to further amplify and ultimately generate additional misogynistic and racist content.

Compounding this, 2025 saw a significant loss of credible and well-curated evidence on gender and marginalized communities with the shutdown of a major bilateral development organization, the US Agency for International Development, and the disappearance of decades of evidence housed in its Development Experience Clearinghouse. Ironically, in 2024, the agency was beginning to embrace the use of AI to help make sense of this vast repository of reports. As a recent article on this experiment points out, the learning was promising, but with the loss of the underlying reports, the AI itself is irrelevant. The hemorrhaging of development aid also affects future credible, gender-disaggregated data collection with the threat to national Demographic and Household Surveys.

The hemorrhaging of development aid also affects future credible, gender-disaggregated data collection with the threat to national Demographic and Household Surveys.

So what are we left with? Child protection groups are documenting an exponential year-on-year rise in AI-generated child sexual abuse material. The sheer volume of this content and the apparent lack of any means to stop its proliferation is its own problem. A second-order problem is the algorithmic amplification of violent, misogynist content, replacing more benign content and feeding a dangerous loop generating demand for yet more violence and misogyny. Worse yet, the rise in generative AI may mean the real, replicable evidence we produced is not just forgotten but replaced with artificial and all too often simply fake supposed source material (“AI slop”). As generative AI tools play a bigger role in how people access and summarize information, made-up content, fake citations, and stripped-down summaries relying on a narrow (and already exclusionary) range of actual primary sources can start to stand in for rigorous intersectional feminist research—especially when reliable sources are scattered, poorly indexed, or no longer maintained. New chokepoints and a very narrow set of gatekeepers eventually will lead to the same type of homogenization and monocropping we’ve seen with capital concentration in other sectors.

Safeguarding feminist knowledge

These failures matter most in areas like sexual and reproductive health and rights, gender-based violence, and the rights and inclusion of LGBTQI+ and other marginalized communities. Without intentional intervention, AI does not just leave evidence out—it can replace it, reshaping public understanding in ways that advantage backlash movements and authoritarian storytelling. This is not a neutral distribution of power, and not merely a ‘manosphere’; we need to recognize and name the violence this label represents.

As others have argued, we need new models for collectively stewarded data, including models establishing data commons. It is critical that we rebalance these systems by ensuring intersectional feminist scholarship is fully represented and that our research is visible and discoverable. This means not only finding ways to digitize material, but being intentional in linking these efforts to alternatives to the current search architecture. We need to create and position strategic online assets, curate our resources, and develop our own ways to optimize their availability in information ecosystems. We must treat search engines and generative AI as shared, contested ground—not neutral tools. We must work together to test what changes outcomes, what helps feminist evidence surface, and what allows false or reactionary content to dominate.

And in the long term, we must seed an alternative to the manosphere – a femisphere, an ecosystem of feminist narratives, evidence, and cultural production to ensure we pass down wisdom as we always have – by sharing our stories to provide a powerful alternative epistemological ecosystem for future generations.

We still have the means to act. First and immediately, our defensive strategy must safeguard feminist knowledge by advocating for the digitization and preservation of critical research, histories, and lessons with particular attention to knowledge that has shaped—or could shape—public and policy understanding. On the proactive side, those who are setting up alternative AI-driven search and synthesis tools must be particularly sensitive to the need to amplify the discoverability of feminist evidence within contemporary information ecosystems. And in the long term, we must seed an alternative to the manosphere: a femisphere, an ecosystem of feminist narratives, evidence, and cultural production to ensure we pass down wisdom as we always have: by sharing our stories to provide a powerful alternative epistemological ecosystem for future generations.

*While AI is commonly understood to refer to algorithmic inference as “intelligence,” this essay rejects the anthropomorphizing of what is simply an extremely rapid example of the infinite monkey theorem at work (noting also that monkeys, like most living species, exercise an actual intelligence that goes well beyond trial-and-error).