Federated AI:
Collaboration with Ownership

The infrastructure rebuilding trust between creators, AIs, and users.

The next AI breakthrough requires unprecedented cooperation, not value extraction.

A protocol that brings your data to AI products and users, while respecting your terms, with creditation and compensation flowing back to you.

Why Data Owners and Publishers Hold the Key to AI's Future

AI today - running out of data

AI's progress is slowing: not because we've reached the limits of compute, but of publicly available data.

The most valuable and trustworthy information in the world — from science, medicine, research, finance, and journalism — remains out of reach. While some data owners have never participated in AI training, an increasing number are now actively opting out altogether.

Why AI can't progress beyond?

The information market is built on broken incentives that penalize sharing and reward exclusivity:

Centralized architectures: When AI copies and trains on data, it erases ownership and can expose contributors to privacy and IP risks.

No attribution or control: As ownership and intent is erased, attribution, compensation and control of one's data is lost.

This makes participation unsustainable at scale.

(left) The centralised architecture: AI as an intermediary between those who have information and those who seek insights.
(right) Attribution-based control: AI as a communication mechanisms that fetches the insights from the data owners, reasons and provides the output to the end-user.

Attribution-Based Control Graphic

Missing Link: Attribution-Based Control (ABC)

ABC is a paradigm shift that aims to reconnect those who seek insights with those who hold them — directly, without intermediation. This is beneficial for all stakeholders:

  • Data owners decide which predictions they participate in (usage terms, compensation, privacy).
  • AI developers gain access to massive amount of high-quality, compliant, trustworthy data to advance AI.
  • End-users decide which sources they trust and consult, with proper attribution and transparency.

A Technical Framework for ABC

We are building a networked protocol that facilitates Attribution-Based Control. It consists of:

  • Syft Protocol & Tools for AI participation: Enable data owners to share data under enforceable terms for AI, while AI developers and AI users can access them in a fully traceable way through APIs and MCP connectors to the network.
  • Federated AI Network: The network facilitated by Syft where participants can discover peers, exchange queries, and build state-of-the-art, network-sourced AI.

👤 👤 👤 🏢 🏢 🏢 👥 Federated AI Network End-users, AI Providers discover & query the networked intelligence. Router Router Router Router 🚪 Syft Routers Gateways to data & models that enforce attribution-based control. Datasite Datasite Datasite Datasite Datasite Datasite 📦 SyftBox Protocol Secure, E2EE, permissioned transport and storage. Siloed Data & Models Distributed, private, on any platform

The Road Ahead: Federating AI & Collaboration

We believe this can transform how we access and use intelligence across every domain, with massive impact on scientific discovery, the next AI frontier, and user experience.

Enhanced User Experience

AI users get informed from the sources they trust

Accelerated Discovery

Scientific discovery is accelerated through private data access

AI Scale

Unlocks orders of magnitude more data and compute by aligning incentives for participation

Trustworthy AI

Embeds privacy, attribution, and auditability into every query

Accessibility

Anyone can leverage, upon respecting the terms, expert data in any field

Fairness

Any data contributor can be compensated for the information they bring

If you want to explore or have critiques, please get in touch - we would love to hear from you!

Ready to implement ABC?

Join us in building the future of collaborative AI where data owners, developers, and users all benefit.