Feminist AI principles are not scarce. What has been scarce is a repeatable process for getting from a principle to a deployed system that someone actually uses.
Most technology projects for civil society start at the tool and work backwards to a justification. Ours starts at the problem, forces it to be articulated in writing, and only then asks whether AI is the right instrument at all. A significant part of the method’s value is that it lets organizations reach the answer “not this, not yet” cheaply, before money and reputation are committed.
Define the problem, the affected communities, the human rights risks, and the theory of change.
This phase begins with structured learning: modules on feminist AI, data, models, and impact assessment. Participants then write a concept paper articulating the problem and the proposed intervention, with peer mentor feedback and discussion across the network. Papers are shared at regional convenings.
No funding attaches to this phase, by design. The paper is the filter.
Build and test an early AI-enabled tool, dataset, method, or governance intervention, with expert mentorship.
Organizations work in the AI Innovation Sandbox with a dedicated technical mentor, moving through design, testing, feedback, and refinement toward a minimum viable product. The mentor handles technical complexity; the organization retains ownership of the direction and the output. Structured protocols gather end-user feedback, with attention to how experience differs across users.
Seed funding attaches to this phase.
Validate the solution in context, document what is learned, and assess whether it can responsibly be scaled.
The tool is deployed in the community it was built for. Feedback is gathered and folded back in. Safety, effectiveness, and actual benefit are assessed against baseline indicators set at the outset — not against adoption metrics alone.
Seed funding attaches to this phase.
Assessment against existing international human rights law, using frameworks including HUDERIA, rather than bespoke ethics criteria invented for the occasion.
Organizations own their data, their code, and their outputs. Technical partners support; they do not acquire.
Process, decisions, and failures are documented from the start, so the method itself can be replicated without us.
Each prototype is examined for problem-solution fit with the target community, replication potential, post-project viability, and measurable impact against baseline.
The method draws on the research practice built through the Feminist AI Research network between 2021 and 2024, and is now operating at scale through the Gender & AI Innovation Collective.