Today, Ritual and Nillion are excited to announce an ongoing partnership that will enable privacy-preserving model inference and storage on the Ritual Network.

New Era of Privacy for Decentralized AI

Ritual is building the first decentralized execution layer for AI. Ritual’s Chain will feature a custom VM optimized for AI-native operations across models. Our first product, Infernet, is already live and enables developers on any EVM-chain to access models from our node network via smart contracts. You can read more on our product page here.

Nillion’s “blind” computation technology makes use of multi-party computation (MPC) gadgets to enable the storing of sensitive and private data along with the ability to compute over this data without revealing it to third-parties.

This extends well into the world of AI/ML. Many users may have model inputs that they would like hidden from parties hosting and serving the model. MPC can help split the model input among multiple parties such that no single party is aware of the model input, yet still computes something over their share of data that is useful for generating the end inference.

Symmetrically, it may be the case that the creator of a model wants to enable others to run their model without necessarily divulging the model weights. Typically, running a proprietary model means either in-housing the necessary compute or trusting a third-party to not leak model weights. With MPC, model execution can be federated across multiple parties, such that it’s difficult to reconstruct the original model aside from collusion, enabling new possibilities for model creators.

Ritual has partnered with Nillion to make this a reality, launching a joint integration that will enable blind (privacy-preserving) model inference and storage on the Ritual network.

Private Data, Public Benefit

Enabling both blind inference and model storage over the Ritual network with Nillion offers several key improvements for users and builders:

Enhanced user data protection: Applications that previously wanted to build models that can interface with private data (i.e. ingesting proprietary market data as an input, or utilizing personal consumer information for social or financial use cases) are now enabled, without needing to jump through hoops. While there are other solutions like FHE that can enable similar properties, they aren’t computationally feasible for large-scale productionized AI/ML use cases. MPC scales today with what users actually want out of their models.

Secure model sharing: Thanks to MPC, proprietary models can be run by external third parties without leakage of model weights and other private model information. This unlocks a new paradigm; previously, creators of custom models would need to run the models themselves or through a trusted third party in order to make them accessible to users and monetize. By eliminating trust assumptions via MPC, model creators on Ritual can have Ritual nodes run their models while still preserving their valuable IP.

Accelerated enterprise adoption: A huge hurdle that larger, institutional consumers of AI models typically contend with are privacy concerns around internal and user data. With Ritual and Nillion, enterprises can access external AI/ML models while respecting their own security needs.

Integrating Blind Computation into Ritual

Nillion’s blind computation network will be integrated to provide dedicated resources for (1) generating inferences over private data and (2) storing private models on the Ritual.

Specifically, this entails:

1. Dedicated cluster creation: Nillion will establish a specialized cluster dedicated to handling model storage and inference for the Ritual network.
2. On-chain contract monitoring: Nodes within that cluster will continuously monitor smart contracts on Ritual for incoming blind inference requests.
3. Client integration: Nillion’s client will be integrated as a sidecar on the RitualVM, so developers and users on Ritual Chain will have native access to private inference.
4. Output reconstruction: The results of blind inference conducted by the Nillion network will be easily reconstructed by Ritual’s end users, ensuring no data leakage end-to-end.

Privacy-first Ritual Applications

Working with Nillion opens up a novel design space for secure, privacy-first applications across blockchain and AI. Some of the use cases include:

• Private On-chain Mechanisms: With Ritual and Nillion, core on-chain primitives can be reimagined under the lens of privacy. From orderbooks where user bids are kept private in a dark-pool setting, to enabling AI agents to interact with prediction markets without leaking their strategy, the obfuscation enabled by MPC can help reimagine the game-theoretic dynamics that play out with on-chain applications today.
• Private Retrieval Augmented Generation (RAG): RAG based systems are critical in providing LLMs with contextual information from vector databases, enabling agentic infrastructure and lowering model hallucination. MPC enables private RAG on Ritual, allowing vector databases to be queried without information leakage.
• Trustless Identity: Anonymization and pseudonymization layers can be largely enhanced by the usage of private consumer data and machine-learning based identifiers. With Nillion, users on Ritual can use network models to construct new identity primitives without having to actually divulge private information used as inputs.

These potential applications are just the tip of the spear. We’re calling on developers, researchers, and visionaries across all sectors to help us explore this new frontier. If you have a novel idea for building with Ritual and Nillion at the intersection of privacy, crypto and AI, make sure to reach out to us or submit your idea to the Ritual Altar program.

Looking Forward

Our partnership with Nillion is delivering the solution to a long standing need: privacy-preserving machine learning. Unlike many other cryptographic or probabilistic primitives, MPC scales today and provides developers and users with the ability to keep both inputs/outputs AND models private, which unlocks new functionality for both builders and users.

We’re excited to work with the community to explore the full potential of this integration. We’ll be releasing more POCs over the coming weeks to demonstrate how to implement Nilion blind computing in projects built on Ritual soon, so keep an eye out.

For all future updates, follow Ritual and Nillion on X. Got ideas for Ritual and Nillion’s tech? Submit them to Ritual Altar, or join our Discord community and we’ll be in touch.