Data sovereignty for coastal AI, in five layers.
When AI arrives on the coast, it should arrive on community terms.
For decades, coastal data has followed an extractive pattern: knowledge flows outward, models are trained elsewhere, and results return as products or reports. Benthos is built around a different protocol, where communities actively shape the datasets that describe them.
The people who collect the data decide what leaves the territory. This is a clause in the partnership contract and a constraint in the code, not just an intention. Funders and research partners receive agreed aggregates; raw data stays where it was spoken. Sovereignty lives in the decisions we encode: in how the database is structured, who can access what, and how every query is logged. This is where a promise becomes enforceable.
“Sovereignty lives in the schema.
In the structure of the data itself.”
Field organisations lead the relationship with community. Research institutes shape the methodology. Funders enable the work. Benthos builds the technical layer that carries it. Each role has clear boundaries, ensuring communities retain control while research and funding scale effectively.
Five layers, from field collection to funder reporting. Each one is a place where the line between what stays and what leaves gets drawn.
This is the architecture we are building.
Community chooses what to log, and who logs it. Consent is granular and explicit.
Raw data sits on infrastructure owned or controlled by the community. Data remains strictly within the territory unless explicit consent is granted.
Named keys and a readable audit trail. Access can be revoked in the system itself, instead of relying on informal email requests.
Only agreed indicators cross the territorial boundary into external systems. Aggregates that protect individual identity, using differential privacy where needed.
Reports are agreed exports. Communities see everything that leaves.
This architecture introduces friction. Onboarding is slower, funders receive aggregates instead of raw data, and communities need technical capacity to exercise the rights it preserves. A sovereignty that costs nothing protects nothing.
Sovereignty is an architectural choice about who a system serves, and the discipline to keep that choice legible in the code. That is what Benthos builds.
This thesis sits in conversation with the CARE Principles for Indigenous Data Governance (Carroll, Garba, Figueroa-Rodríguez et al., 2020), which name the social commitments of community data: collective benefit, authority to control, responsibility, ethics. We read CARE through an engineering question: where does each commitment land in the code?
It also responds to the literature on data colonialism (Ricaurte, 2019), and sits alongside the FAIR Principles (Wilkinson et al., 2016) where the two frameworks move in different directions. Where layer 04 invokes differential privacy, the underlying mathematical work is Cynthia Dwork's.
If you see an opportunity to refine a layer, or know a coastal
project that needs this architecture, reach out directly.