# Overview

### Background

As decentralized AI Agents become foundational components of the next-generation Web3 infrastructure, their secure, transparent, and verifiable execution is critical. Unlike centralized systems, Web3’s openness introduces new risks:

* **Ownership Ambiguity:** Difficult to prove data/model ownership on-chain.
* **Authenticity Risks:** Agent behavior and output cannot be trusted by default.
* **Lack of Auditability:** No provable link between the used data, the model executed, and the output produced.
* **No On-Chain Verifiability:** Inference results can be fabricated without cryptographic accountability.

In LazAI, each iDAO manages its own data, model, and agent workflows. While this empowers decentralized autonomy, it also raises the need for a trust-minimized, cryptographically verifiable execution framework.

### iDAO’s Key Security & Privacy Challenges

1. **Data Misuse & Privacy Breach**
   1. Sensitive datasets may be exposed or misused during AI processing.
   2. Users lack visibility into whether their inputs were securely processed.
2. **Unverifiable Inference**
   1. AI inference is typically executed off-chain or locally; users cannot validate results or trace data provenance.
   2. The same model may be reused, modified, or forked by different iDAOs without traceability.
3. **Unclear Attribution & Revenue Distribution**
   1. Models often rely on multiple data sources; without execution proofs, revenue sharing becomes opaque.
4. **Untrusted Agent Runtime**
   1. Many AI Agents run on centralized, uncontrolled hardware—susceptible to tampering or malicious behavior.

### LazAI’s Solution: TEE-First, ZK-Optional, OP-Compatible Verified Execution

LazAI introduces a Verified Computing Framework built on three composable execution trust models:

<table data-header-hidden><thead><tr><th width="119.921875"></th><th width="109.5625"></th><th width="305.90625"></th><th width="196.3046875"></th></tr></thead><tbody><tr><td><strong>Mode</strong></td><td><strong>Primary Usage</strong></td><td><strong>Strength</strong></td><td><strong>When to Use</strong></td></tr><tr><td>TEE Mode</td><td>Default</td><td>Trusted execution environment (Intel TDX and SGX are both OK) with remote attestation</td><td>Most inference and fine-tune tasks</td></tr><tr><td>TEE + ZK Mode</td><td>Optional extension</td><td>Adds zero-knowledge proofs for selective inputs, outputs, or logic constraints</td><td>Privacy-sensitive or regulatory tasks</td></tr><tr><td>Optimistic Mode (OP)</td><td>Backup</td><td>Enables fraud-proof-based verification when TEE is unavailable</td><td>Lightweight agents or fallback arbitration</td></tr></tbody></table>

### When to Use ZK or OP

ZK and OP serve as optional augmentation to TEE-based computing:

**Use ZK when:**

* Inputs or outputs must be kept private (e.g., mental health input).
* Output must satisfy constraints (e.g., model temperature < 0.9).
* Regulatory compliance requires provable logic.

**Use OP when:**

* Either TEE is not used, or trust assumptions are weaker.
* Third-party challengers need to verify computation after execution.
* Arbitration or slashing is necessary (e.g., data misuse, forged proofs).

### Cryptographic Proof Models

<table data-header-hidden><thead><tr><th width="124.69921875"></th><th width="284.921875"></th><th width="458.4375"></th></tr></thead><tbody><tr><td><strong>Asset Type</strong></td><td><strong>Proof Type</strong></td><td><strong>Purpose</strong></td></tr><tr><td>Dataset</td><td>Merkle Root + Provenance Hash</td><td>Anchors original data, prevents replacement</td></tr><tr><td>Model</td><td>TEE Attestation + Param Hash</td><td>Verifies model version and source</td></tr><tr><td>Inference</td><td>TEE Signature (optional ZK)</td><td>Verifiable output, optionally private</td></tr></tbody></table>

### Integration with LazAI Infrastructure

**Quorum-Based Consensus + VSC Coordination**

* Quorum Validators stake on LazChain and form trust domains for iDAO.
* VSC (Verifiable Service Coordinator) aggregates proofs from TEE nodes and submits them on-chain.
* Verifier Contract on LazChain checks TEE signatures, ZK proofs, or OP dispute proofs.

**Challenger Mechanism (for OP Mode)**

* Elected challengers monitor inference results and data-model linkage.
* Upon detecting fraud, challengers submit Fraud Proofs.
* Quorum enforces Slashing, penalizing iDAOs via token burn or DAT reduction.

### Summary

LazAI’s Verified Computing architecture provides a hybrid, multi-layered trust framework tailored for decentralized AI ecosystems. With a TEE-first, ZK-assisted, and OP-compatible design, it addresses key pain points in:

* Executing private, auditable AI inference tasks
* Enforcing data/model provenance
* Validating AI agent behavior at low cost
* Coordinating decentralized trust via Quorum and VSC

This framework transforms iDAOs from isolated compute units into co-verifiable AI entities, anchored by programmable security, fine-grained delegation, and decentralized validation.

LazAI’s architecture lays the groundwork for a future where AI is both trustless and transparent, and every computation, dataset, and model update can be proven, traced, and monetized across on-chain and off-chain domains.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.lazai.network/verified-computing-framework/overview.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.