[Part 3/3] Verida Technical Litepaper: Self-Sovereign Confidential Compute Network to Secure Private AI
[Part 3/3] Verida Technical Litepaper: Self-Sovereign Confidential Compute Network to Secure Private AI
This is the third and final post to release the “Verida Technical Litepaper: Self-Sovereign Confidential Compute Network to Secure Private AI”. Read Part 1 and Part 2.
Confidential Compute Nodes
Confidential Compute Nodes running on the Verida Self-Sovereign Compute Network operate a web server within a secure enclave environment to handle compute requests and responses.
There will be different types of nodes (i.e., LLM, User API) that will have different code running on them depending on the service(s) they are providing.
For maximum flexibility, advanced users and developers will be able to run compute nodes locally, on any type of hardware.
Nodes have key requirements they must adhere to:
GPU access is required for some compute nodes (i.e., LLM nodes), but not others. As such, the hardware requirements for each node will depend on the type of compute services running on the node.
Code Verifiability is critical to ensure trust in the compute and security of user data. Nodes must be able to attest the code they are running has not been tampered with.
Upgradability is essential to keep nodes current with the latest software versions, security fixes and other patches. Coordination is required to ensure applications can ensure their code is running on the latest node versions.
API endpoints are the entry point for communicating with nodes. It’s essential a web server host operates within the secure enclave to communicate with the outside world.
SSL termination must occur within the secure enclave to ensure the host machine can’t access API requests and responses.
Resource restraints will exist on each node (i.e., CPU, memory) that will limit the number of active requests they can handle. The network and nodes will need to coordinate this to ensure nodes are selected that have sufficient resources available to meet any given request.
Interoperability and Extensibility
In order to create an efficient and highly interoperable ecosystem of self-sovereign API’s, it's necessary to have a set of common data standards. Verida’s self-sovereign database storage network provides this necessary infrastructure via guaranteed data schemas within encrypted datasets, providing a solid foundation for data interoperability.
Developers can build new self-sovereign compute services that can be deployed on the network and then used by other services. This provides an extensible ecosystem of API’s that can all communicate with each other to deliver highly complex solutions for end users.
Figure 4: Interoperable data between self-sovereign AI services
Over time, we expect a marketplace of private AI products, services and APIs to evolve.
Service Discovery
Verida’s self-sovereign compute network will enable infrastructure operators to deploy and register a node of a particular service type. When an API needs to send a request to one of those service types, it can perform a “service lookup” on the Verida network to identify a suitable trusted, verifiable node it can use to send requests of the required service type.
User Data Security Guarantees
It is essential to protect user privacy within the ecosystem and prevent user data leaking to non-confidential compute services outside the network. Each service deployed to the network will be running verifiable code, running on verifiable confidential compute infrastructure.
In addition, each service will only communicate with other self-sovereign compute services. Each API request to another self-sovereign compute service will be signed and verified to have been transmitted by another node within the self-sovereign network.
Tokenized Payment
The VDA token will be used for payment to access self-sovereign compute services. A more detailed economic model will be provided, however the following key principles are expected to apply.
End users will pay on a “per-request” basis to send confidential queries to compute nodes and the services they operate. The cost per request will be calculated in a standardized fashion that balances the computation power of a node, memory usage and request time. Applications can sponsor the request fees on behalf of the user and then charge a subscription fee to cover the cost, plus profit, much like a traditional SaaS model.
Node operators will be compensated for providing the confidential compute infrastructure to Verida’s Self-Sovereign Compute Network.
Builders of services (i.e., AI Prompts and Agents) will be able to set an additional fee for using their compute services, above and beyond the underlying “per-request” compute cost. This open marketplace for AI Agents and other tools drives innovation and provides a seamless way for developers to generate revenue from the use of their intellectual property.
Verida Network will charge a small protocol fee (similar to a blockchain gas fee) on compute fees.
Other Use Cases
Data Training Marketplaces
Verida’s Private Data Bridge allows users to reclaim their private data from platforms such as Meta, Google, X, email, LinkedIn, Strava, and much more.
Users on the Verida network could push their personal data into a confidential compute service that anonymizes their data (or generates synthetic data) which is made available to various AI data marketplaces. This provides an option for users to monetize their data, without risk of data leakage, while unlocking highly valuable and unique datasets such as private messages, financial records, emails, healthcare data for training purposes.
Managed Crypto Wallets
There is a vast array of managed wallet services available today that offer different trade-offs between user experience and security.
Having an always available cloud service that can protect user’s private keys, but still provide multiple authorization methods for a user is extremely useful to onboard new users and provide additional backup protection measures for existing users.
Such a managed wallet service becomes rather trivial to build and deploy on the Verida self-sovereign compute network.
Verifiable Credentials
Verida has extensive experience working with decentralized identity and verifiable credential technology, in combination with many ecosystem partners.
There is a significant pain point in the industry, whereby developers within credential ecosystems are required to integrate many disparate developer SDK’s to offer an end-to-end solution. This is due to the self-sovereign nature of credentials and identity solutions where a private key must be retained on end user devices to facilitate end-to-end security.
Verida’s self-sovereign compute network can provide a viable alternative, whereby application developers can replace complex SDK integrations with simple self-sovereign APIs. This makes integration into mobile applications (such as identity wallets) and traditional web applications much easier, simpler and viable.
This could be used to provide simple API integrations to enable:
Identity wallets to obtain access to a user’s verifiable credentials
End users to pre-commit selective disclosure rules for third party applications or identity wallets, without disclosing their actual credentials
Provide trusted, verifiable universal resolvers
Trust registry APIs
Any complex SDK that requires a user’s private key to operate, could be deployed as a micro service on Verida’s self-sovereign compute network to provide a simpler integration and better user experience.
Conclusion
Verida's mission to empower individuals with control over their data continues to drive our innovations as we advance our infrastructure. This Litepaper outlines how the Verida Network is evolving from decentralized, privacy-preserving databases to include decentralized, privacy-preserving compute capabilities, addressing critical issues in AI data management and introducing valuable new use cases for user-controlled data.
As AI faces mounting challenges with data quality, privacy, and transparency, Verida is at the forefront of addressing these issues. By expanding our network to support privacy-preserving compute, we enable the more effective safeguarding of private data while allowing it to be securely shared with with leading AI models. This approach ensures end-to-end privacy and opens the door to hyper-personalized and secure AI experiences.
Our solution addresses three fundamental problems: enabling user access to their private data, providing secure storage and sharing, and ensuring confidential computation. Verida's "Private Data Bridge" allows users to securely reclaim and manage their data from various platforms and facilite its use in personalized AI applications without compromising privacy.
While we are not focusing on decentralized AI model training or distributed inference, Verida is committed to offering high-performance, secure, and trusted infrastructure for managing private data. We are collaborating with partners developing private AI agents, AI data marketplaces, and other privacy-centric AI solutions, paving the way for a more secure and private future in AI. This empowers users to be confident about the ways their data is used, and receive compensation when they do choose to share elements of their personal data.
You can learn more or get involved at https://www.verida.network/