NEAR DCML explained: confidential, verifiable and monetizable AI via TEEs
Quote406 words
Market since$NEAR+64%since this pieceHow it played out ↓Explains Decentralized Confidential Machine Learning: models run in Confidential VMs inside TEEs (Intel TDX, AMD SEV, NVIDIA H100+), with encrypted weights, payment channels and Proof of Response. This enables monetizable open models, private user data and verifiable inference.
Key claims
Model weights decrypt only inside the CVM, which enforces payment
Proof of Response guarantees a reply or cryptographic proof the provider is offline
Verifying model hash inside the TEE prevents model substitution
WTF is Decentralized Confidential Machine Learning?
@NEARProtocol's DCML is a decentralized system for creating, deploying, and using AI models and agents. It is designed to be a confidential, decentralized AI cloud that allows models to be both open-source and monetizable, private and verifiable, while preserving user ownership of data.
➀ How It Works
➤ Trusted Execution Environments (TEEs): The system's foundation is the use of TEEs, which are secure enclaves available in modern CPUs (Intel TDX, AMD SEV) and GPUs (NVIDIA H100 and newer). TEEs allow code to run on third-party hardware in a way that the hardware owner cannot see or tamper with the user's data or the computation itself.
➤ Confidential Virtual Machines (CVMs): It uses an abstraction called a CVM, where a special Docker container runs inside the TEE. This ensures that only the intended code processes user requests and that all data remains private from the hardware operator and even the CVM's creators.
➤ To enable monetization, a model creator can encrypt their model's weights. These weights can only be decrypted and used inside the CVM, which enforces payment for inference via smart contract-based payment channels.
➤ Network reliability is ensured by a mechanism called Proof of Response. This guarantees that a user will either receive a response from a service provider or get cryptographic proof that the provider is offline, allowing the user to avoid paying for unrendered services.
➁ Why It Matters
➤ Solves Monetization for Open Models: It creates a sustainable business model for open-source AI, which currently lacks one. Devs can distribute models widely without losing the ability to get paid for their usage.
➤ Guarantees User Privacy: Unlike current advanced AI models that require users to give up their data privacy , DCML makes it so neither the hardware host nor the model developer can see a user's private data, inputs, or outputs.
➤ Enables Verifiable Inference: Users can be cryptographically certain that the AI model they intended to use was the one that actually ran, preventing providers from substituting a smaller or different model. This is achieved by verifying the model's hash inside the TEE.
➤ Allows for Community-Owned AI: The framework allows multiple, distrusting parties to pool funds to train a new model. The model is trained and encrypted entirely within TEEs, so no single contributor can access the unencrypted weights and all participants can share in the profits programmatically.
AI is $NEAR anon.
DYOR.
QUOTING @NEARProtocol
What is DCML? Decentralized Confidential Machine Learning enables any party that needs a reliable, confidential AI workload to leverage another party’s compute securely. It provides the foundation for private, verifiable and user‑owned access and monetization of AI models 👇
What is DCML? Decentralized Confidential Machine Learning enables any party that needs a reliable, confidential AI workload to leverage another party’s compute securely. It provides the foundation for private, verifiable and user‑owned access and monetization of AI models 👇
1 image on X
How the covered assets moved · since 22 Jul 2025
Prices as of 23 Oct 2026This explainer walks through how things work without making a call on the token, so price moves since publication are shown as market context.
Covered in this piece
Since then
Zero has written 3 more pieces on NEAR Protocol and $NEAR since this one.
Daily closes in USD from DefiLlama and Yahoo Finance. Only a bullish or bearish thesis is read as a call; explainers, maps and mentions are context. Not financial advice.