📄Phala Network

• 👤 Author: [20200167 / Kim Jinhwan]
• 📆 Presentation Date: [YYYY-MM-DD]

---

1. Overview

• Project Name:
• Category: [e.g., Decentralized AI Infrastructure, AI Payment Rails, etc.]
• Key Technologies / Platforms: [e.g., Solana, Bittensor, Sahara AI, ZKML, etc.]
• Official Links:
• Website
• Docs
• GitHub
• X
• Discord

📌 Summary

Phala Network is a privacy-preserving AI infrastructure that enables secure data processing. It combines blockchain with Trusted Execution Environments (TEEs) to ensure decentralized, verifiable AI execution. User data remains encrypted, and all operations are logged immutably on-chain. The project is already live with real AI agents and dApps in production. Phala offers a compelling Web3 solution to trust, privacy, and governance in AI.

---

2. Background & Problem Statement

• What real-world problem does this project aim to solve?
  Phala Network addresses a fundamental issue in modern AI infrastructure: the centralization of computation and data control, and the lack of robust privacy guarantees. Today’s AI services are primarily operated by a few cloud giants (e.g., AWS, Google Cloud), which leads to:

  • Monopolized control over data and models
  • Privacy risks for user data
  • High costs and lack of transparency
  • Potential censorship or restriction by governments or corporations

  As AI becomes critical to more aspects of society, issues such as data ownership, trust in computation, and equitable access to AI services are becoming increasingly important.

• What are the limitations of existing (centralized) approaches?
  Centralized AI infrastructure has structural limitations:

  • Censorship and control risks: Corporations or governments can unilaterally restrict access to certain models or datasets
  • Data exposure threats: Central servers present high risks of hacking or misuse of personal data
  • Lack of transparency: Users cannot verify how models are run or what data is used
  • Cost barriers and monopolies: Only large companies can afford to operate or use enterprise-level AI infrastructure
  • Slower innovation: Centralized control limits experimentation with open-source or alternative AI models

• Why is this problem especially relevant in the context of AI? AI relies on massive computing power and access to sensitive data, making trust, transparency, and privacy essential. But centralized systems create friction in all three areas:

  • Black-box behavior: Internal processes are opaque and unverifiable
  • Weak privacy protection: Input data can be leaked or exposed
  • Limited accessibility: Open participation in AI development is restricted

  Phala addresses these challenges by: - Using TEE (Trusted Execution Environments) to securely execute AI workloads - Enabling verifiable outputs via zkCompute and on-chain proofs - Creating an open, decentralized AI ecosystem where computation is transparent and censorship-resistant

  In a world where AI is reshaping economies, governance, and daily life, these problems—and their solutions—are becoming more critical than ever.

• 해결하려는 실제 문제 현실 세계에서 Phala Network가 주목하는 주요 문제는 AI 연산과 데이터 처리의 중앙화 및 프라이버시 보호의 구조적 한계입니다. 현재 대부분의 AI 서비스는 AWS, Google Cloud 등 거대 중앙 클라우드 사업자에 의해 운영되고 있어 다음과 같은 문제가 발생합니다:

  • 데이터와 AI 모델에 대한 독점적 통제
  • 사용자 데이터 프라이버시 침해 위험
  • 높은 비용과 낮은 투명성
  • 특정 정부나 기관에 의한 검열 및 접근 제한 가능성

  AI가 점점 더 많은 영역에서 핵심 인프라로 자리 잡으면서, 데이터 소유권, 결과의 신뢰성, AI 접근의 민주화 문제가 점점 더 중요해지고 있습니다.

• 기존(중앙화) 방식의 한계 중앙 집중형 AI 인프라는 다음과 같은 구조적 한계를 가집니다:

  • 검열 및 통제 가능성: 특정 기업이나 정부가 AI 모델 또는 데이터 접근을 임의로 제한하거나 차단할 수 있음
  • 데이터 노출 위험: 중앙 서버에서 데이터가 운영되며, 해킹·오남용 등 보안 위협 존재
  • 투명성/신뢰성 부족: AI 모델이 어떻게 작동하는지, 어떤 데이터를 기반으로 연산되었는지 외부에서 검증 불가능
  • 비용 및 독점 구조: 고비용 인프라 운영으로 인해 소수 대기업만 시장 진입 가능
  • 혁신 속도 저하: 중앙화된 통제 구조는 실험적 AI 모델 개발과 오픈소스 혁신을 저해함

• AI 맥락에서의 문제 중요성 AI는 막대한 연산력과 민감한 데이터를 기반으로 동작하기 때문에, 신뢰성, 투명성, 프라이버시가 모두 중요한 요소입니다. 그러나 중앙화된 시스템은 다음과 같은 제약이 존재합니다:

  • AI의 ‘블랙박스’ 문제: 내부 구조가 불투명하고 외부 검증이 어려움
  • 프라이버시 보호 부족: AI 모델에 입력되는 데이터가 외부에 노출될 가능성
  • 공정한 접근성 부족: 누구나 AI를 활용하고 개발할 수 있는 환경이 부족함

  Phala Network는 이러한 문제를 해결하기 위해, - 신뢰 실행 환경(TEE)을 활용해 데이터를 보호하며 - zkCompute를 통해 연산 결과를 온체인에서 검증하고 - 분산형 AI 생태계를 구축하여 AI의 자유로운 접근성과 투명성을 확보하려고 합니다.

  이는 AI가 사회 전반에 미치는 영향력이 커지는 오늘날, 반드시 해결해야 할 핵심 과제입니다.

---

3. How It Works

🔍 3.1 Project Approach

• How does this project intend to solve the problem defined in Section 2? Phala Network combines decentralized AI infrastructure with Trusted Execution Environments (TEE) to ensure privacy and trust in AI execution and data handling. Key strategies include:

  • Using TEE hardware (e.g., Intel SGX) to securely process AI tasks and sensitive data on a distributed network
  • Enabling the creation of AI-Agent smart contracts that autonomously operate in natural language across dynamic environments
  • Making all AI results verifiable on-chain, ensuring full transparency and auditability

• What is the core idea behind the solution?
  Phala’s core innovation is to decentralize the entire AI pipeline — from execution and data privacy to economic incentives:

  • Secure computation without exposing sensitive data
  • Dual-layer verification: protected execution within TEE + on-chain zkProof validation
  • An AI-Agent Economy where developers can build, deploy, and monetize AI agents in a permissionless Web3 environment

• What makes this approach different from existing methods?

Feature	Centralized AI	Phala Network
Data Privacy	Low	Very High (TEE-protected execution)
Transparency & Trust	Limited	Very High (on-chain verifiable outputs)
Scalability & Cost	Expensive, Closed	Low-cost, Open participation
AI Agent Deployment	Centralized	Fully decentralized, Open marketplace
Censorship Resistance	Low	Very High (censorship-proof by design)

• 문제 해결 방법 (Project Approach) Phala Network는 탈중앙화된 AI 인프라와 신뢰 실행 환경(TEE) 기술을 결합하여, AI 데이터 처리와 실행 과정에서의 프라이버시 보호와 결과의 신뢰성 확보라는 문제를 해결하고자 합니다. 주요 전략은 다음과 같습니다:

  • TEE 하드웨어(예: Intel SGX)를 활용해 AI 연산 및 데이터 처리를 분산 네트워크 상에서 안전하게 수행
  • AI-Agent 스마트컨트랙트를 통해 자연어 기반 AI 에이전트 생성 및 실행 가능 — 전통적인 스마트컨트랙트의 한계를 넘어 다양한 외부 환경에서 자율 동작
  • AI 연산 결과는 블록체인 기반으로 검증되며 누구나 검증 가능

• 핵심 아이디어 (Core Idea) Phala의 핵심 아이디어는 AI의 실행, 데이터 처리, 보상 구조를 모두 탈중앙화하는 데 있습니다.

  • 민감한 데이터를 외부에 노출하지 않고 안전하게 연산할 수 있는 환경 제공
  • TEE 내부 연산 → zkProof로 결과 증명 → 온체인 제출이라는 이중 검증 구조
  • 누구나 AI 에이전트를 개발해 Web3 인프라와 연동, 수익화 가능한 AI-Agent Economy 구현

• 기존 방식과 차별점 (What Makes It Different?)

구분	기존 중앙화 AI	Phala Network
데이터 프라이버시	낮음	매우 높음 (TEE 기반 암호화 연산)
신뢰성·투명성	제한적	매우 높음 (온체인 검증 가능)
확장성 및 비용	고비용, 폐쇄적	낮은 비용, 개방형 참여 가능
AI 에이전트 배포	중앙 집중형	완전 탈중앙화, 오픈 마켓플레이스
검열 저항성	낮음	매우 높음 (검열 불가 구조)

---

🏗️ 3.2 Architecture

• Provide a high-level overview of the system architecture.

  Phala Network implements a decentralized AI computing infrastructure that combines Trusted Execution Environments (TEEs) with blockchain to ensure maximum privacy and trust for data and model execution. This hybrid design enables secure, verifiable, and transparent AI workloads in a decentralized setting.

• Include a diagram or describe the components and how they are connected.

Component	Description
Phala Blockchain (on-chain)	A Substrate-based blockchain that manages transactions, user identities, governance, and network state
pRuntime (TEE enclave)	Off-chain execution layer based on Intel SGX or AMD SEV; runs AI computations, Phat Contracts, and processes user data securely
Pherry (Bridge Relayer)	Acts as a secure bridge between the blockchain and pRuntime, relaying encrypted messages and proofs
Worker Nodes	TEE-enabled machines distributed across the network that execute AI models and data processing tasks
Gatekeepers	Special nodes responsible for secure key management, access control, and resource distribution within the network
Clients / Users	End-users or developers who interact with Phala via dApps or interfaces to send AI requests and receive results

• Focus on how data flows between users, models, smart contracts, etc.

  Data Flow (Step-by-Step) 1. The user/client sends a transaction (e.g., AI request) to the Phala blockchain 2. The blockchain validates and logs the request, and sends it through Pherry to the pRuntime enclave 3. The pRuntime executes the smart contract (Phat Contract) and runs the AI model using TEE-protected computation 4. Results are sent back through Pherry to the blockchain, where they are verified and logged 5. he user receives the result, and any reward or staking logic is handled on-chain 6. f needed, external APIs or data sources can be securely accessed by pRuntime to support computation

  Key Architectural Advantages - Privacy: All computations are performed inside secure hardware enclaves (TEE), with no data leakage - Trust: Every result is cryptographically verifiable and logged on-chain - Scalability: Distributed worker nodes allow for horizontal scaling - Interoperability: External APIs and Web2 data can be integrated securely - Token Incentives: All rewards and staking are processed trustlessly via smart contracts

---

🎯 3.3 Core Components

• Describe the key components or modules (e.g., data capsule, model registry, token system)

  1. Data Capsule Concept: Encrypted containers designed to securely store and process sensitive data Role: Ensures that user data remains inaccessible to any party outside of the TEE (pRuntime); data is only decrypted and processed within the secure enclave Key Features:

     • Access and usage are controlled via smart contracts
     • Supports compliance and privacy-by-design
     • Enables fine-grained data permissioning and auditability

  2. Model Registry Concept: A decentralized repository for managing and versioning AI models Role: Allows developers to register, identify, license, and track AI models deployed in the network Key Features:

     • Transparent access control and version history
     • Only verified models can be selected for execution
     • All changes and usage are recorded on-chain

  3. Token System (PHA Token) Concept: A crypto-native economic layer powering payments, incentives, and governance Role:Users pay native tokens (e.g., PHA) to request AI services. Node operators, developers, and data providers are rewarded based on their contributions Supports staking, governance, and dynamic resource allocation Key Features:

     • Incentivizes decentralized participation and sustainability
     • Aligns economic rewards with network utility

  4. Phala Blockchain (On-chain Layer) Role: Serves as the core infrastructure for transaction validation, contract execution, rewards distribution, and governance Key Features:

     • Built on Substrate for scalability and customizability
     • Ensures transparency and immutability of all on-chain records

  5. pRuntime (TEE Execution Layer) Role: Executes all AI computations and Phat Contracts inside trusted hardware environments like Intel SGX Key Features:

     • Off-chain yet secure execution
     • Guarantees that sensitive data is never exposed outside the enclave
     • Forms the computational backbone of Phala

  6. Pherry (Bridge Relayer) Role: Facilitates secure and reliable message transmission between the on-chain blockchain and off-chain pRuntime Key Features:

     • Ensures data consistency and integrity across both layers
     • Acts as a trusted bridge between different trust domains

  7. Gatekeepers Role: Handle secure key management, resource distribution, and permission control within the network Key Features:

     • Operate as a collectively trusted management layer
     • Coordinate secure operations across nodes and enclaves

---

🔁 3.4 Workflow Overview

Explain the overall process or user flow in 3–5 steps.

1. User Request & Data Submission The user sends an AI request via dApp or smart contract, optionally attaching encrypted data or linking external APIs.

2. On-chain Logging & Off-chain Delivery The request is recorded on the Phala Blockchain and securely relayed to the off-chain pRuntime via Pherry.

3. TEE-based Secure Computation A Worker Node executes the AI model inside a Trusted Execution Environment (TEE), ensuring data remains private.

4. Result Verification & Return The result is sent back on-chain with proof and made accessible to the user.

5. Token Rewards & Incentives The user pays in PHA tokens; contributors (node operators, model developers, data providers) are rewarded automatically.

(You may include a diagram or link to an official architecture diagram if helpful.)

---

4. Token Economy (if applicable)

Token Overview

Token Name: PHA
Type: Utility / Governance / Staking
Standard: Substrate-based native token (not ERC-20)

Use Cases in the Ecosystem

- Payments: Users pay in PHA to use AI services and interact with Phat Contracts
- Rewards: Contributors (e.g., node operators, model developers) earn PHA for participating
- Staking: Gatekeepers and worker nodes stake PHA as economic security
- Governance: PHA holders can vote on protocol upgrades and parameter changes

Incentive Mechanism

• PHA creates a sustainable ecosystem where key actors are rewarded based on their role and contribution:

Stakeholder	How They Use / Earn the Token
Data Provider	Earns PHA for contributing quality datasets
Model Developer	Receives PHA based on model usage and licensing
TEE Node Operator	Earns PHA by executing AI computations securely in TEE
Gatekeeper	Staked PHA enables them to manage keys and secure the network
User	Pays PHA to access AI services or deploy Phat Contracts

Tokenomics - Supply Model: Inflationary - Emission: Block rewards + service-based rewards - Staking Mechanism: Required for node/gatekeeper participation - Vesting: Early contributors and team allocations are subject to vesting schedules - Burning: No default burn mechanism, but may occur via governance proposals

---

5. Project Status & Plan

Project Status: Live and Evolving Mainnet Launch: - Phala Network is fully operational. It launched on the Ethereum mainnet in January 2025, following prior deployments on Polkadot and testnets.

Recent Expansion: - The release of Phala Network 2.0 introduced the first OP Succinct Layer 2 on Ethereum, signaling continued technical innovation and a move toward cross-chain AI computation.

User Base & Community - Developer Ecosystem: As of March 2025, Phala’s Ambassador Program includes → 2 Head Ambassadors, → 4 Senior Ambassadors, → 15+ active contributors driving education, marketing, and global community growth. - GitHub Activity & Open Source: Phala is fully open source with 278+ public repositories (as of July 2025). Active development includes regular commits, updates, and feature enhancements, reflecting a transparent and engaged contributor base.

Partnerships & Funding - Strategic Collaborations: Partnered with BUIDL_QUESTS 2025, Autonomys, and APRO to expand privacy-first AI and confidential computing. Phala also collaborates with Bitcoin Gold on secure cross-chain bridges and with several DeFi protocols to integrate privacy-preserving features. - Investments & Grants: Received investment from DWF Labs (June 2025), which now supports PHA with liquidity provisioning and market-making services, enhancing token accessibility and market health.

Real-World Use & Adoption (Hype vs. Traction) - Use Cases: Phala powers real dApps in confidential fintech AI, decentralized cloud computing, and Web3 AI agents. For example, the “Newton” confidential AI agent launched on Base chain in 2025, offering practical automation in the fintech sector. - Growing Ecosystem: Support for Phat Contracts continues to grow, enabling more composable, privacy-preserving applications in DeFi, analytics, and AI.

Token (PHA): Listing & Market Activity - Availability: PHA is actively traded on major exchanges (e.g., Binance, KuCoin), with increasing liquidity supported by market makers like DWF Labs. - 2025 Price Activity: Analysts project a wide range ($0.05 to $1.50) for PHA in 2025 due to high volatility in the AI + crypto sector, with price movements tied to major launches and partnerships, not mere speculation. - Utility: PHA powers governance, staking, and reward mechanisms. Contributors (nodes, developers, ambassadors, data providers) are rewarded through decentralized token flows.

Key Takeaways - Not Just Hype: Phala is a live, production-ready Web3 project with real technical delivery, developer traction, and open-source transparency. - Community-Led Growth: A global contributor base and structured Ambassador Program are actively expanding the ecosystem. - PHA Token in Action: PHA is used daily for service payments, governance, and rewards — not just speculation. - Balanced Perspective: While Phala shows strong momentum and real-world deployments, future success still depends on continued developer adoption and execution in a competitive and volatile space.

---

6. User Experience & Hands-on Review (if applicable)

Try to actually use the project if possible — via a demo, public app, testnet, or simulation. This section should capture your experience as a user or a developer, not just as a researcher.

Here are some prompts to help you reflect:

• What features did you explore?
• What did you do step-by-step?
• Was the onboarding intuitive or confusing?
• How did you deal with a crypto wallet or tokens?
• What felt different from traditional (non-blockchain) services?
• What worked well? What didn’t?

You can also include: - Screenshots - Links to testnet/demo activity - Errors or bugs you encountered

---

7. Why Blockchain

1. Why Does This Problem Require Blockchain? Phala tackles problems deeply embedded in centralized AI systems: privacy vulnerabilities, lack of trust, opaque processes, and potential censorship. Blockchain provides the structural capabilities to overcome these issues:
2. Tamper-proof Execution and Logging: Every computation, data access, and AI model interaction is immutably recorded on-chain, preventing any party from secretly modifying histories or results. This transparent execution ledger ensures sensitive data is processed correctly—and can be verified.
3. Trustless Collaboration: Users and developers don’t need to trust any central intermediary. Blockchain enables peer-to-peer interaction, reducing risk of fraud, data abuse, or censorship by centralized authorities.

4. Automated, Enforceable Incentives: Smart contracts ensure that all actors—data providers, model developers, node operators—are automatically compensated according to rules. Such trustless, programmable reward systems are nearly impossible in traditional off-chain setups.

5. What Does Blockchain Enable That Traditional Solutions Cannot? Censorship Resistance: No single actor (e.g., government, company) can block access to models, freeze assets, or remove services arbitrarily. This is critical as AI becomes embedded in finance, communications, and personal infrastructure.

6. Verifiable Processing: Through cryptographic proofs and TEE attestation, users don’t need to take the platform’s word—they get on-chain proof that computation was secure and private.

7. Open Participation & Programmability: Anyone can join the network, deploy models, or build on existing services without permission. Traditional platforms restrict participation to selected partners—blockchain removes that barrier.

8. Transparent Data/Model Usage: Who used which dataset, how the model behaved, and how incentives were distributed—all of it is recorded permanently on-chain. This enhances auditability, regulatory compliance, and scientific reproducibility

9. Is Decentralization Critical in This Use Case? Absolutely. Decentralization is fundamental to the integrity and fairness of Phala’s AI ecosystem:

10. Prevents Monopoly & Abuse: Centralized AI gives disproportionate power over data, access, and governance to a few corporations. Decentralization distributes this control, reducing bias, privacy violations, and monopolistic outcomes.

11. Enables User Sovereignty: With decentralized custody and computation, users can control their data, decide what to share or monetize, and avoid surveillance-driven business models.
12. Improves Resilience: Distributed networks are more resistant to downtime, cyberattacks, or political interference—no single point of failure can halt critical AI services.
13. Fosters Innovation: A decentralized, open ecosystem attracts a diverse global contributor base, encourages experimentation, and accelerates breakthroughs through permissionless, community-driven development.

---

8. Insights & Limitations

✅ Key Takeaways

Deep Privacy Integration By leveraging Trusted Execution Environments (TEEs) for off-chain compute, Phala ensures that sensitive data is only decrypted and processed within secure enclaves—minimizing the risks of leaks or misuse. This architecture is especially powerful for privacy-first AI.

Open, Programmable Infrastructure Anyone can deploy AI agents or services using smart contracts (Phat Contracts), fostering a permissionless innovation environment. This enables rapid iteration and decentralization of AI development.

Verifiable, Trustless Execution All AI-related activities—model calls, data usage, execution results—are logged immutably on-chain, providing full auditability and reducing reliance on trust.

Clear Incentive Alignment via Token Economy The PHA token enables transparent, automated, and on-chain distribution of rewards to data providers, model developers, and node operators—sustaining long-term ecosystem engagement.

Active Community & Real Adoption Phala is not just theoretical. With a live mainnet, real integrations in DeFi and confidential computing, and an active ambassador/dev community, it demonstrates traction beyond hype.

⚠ Limitations / Open Questions

Technical Challenges TEE Limitations - Hardware vulnerabilities (e.g., Intel SGX), limited memory, and reliance on specific chip vendors pose risks to decentralization, trust, and resilience.

Performance Overheads - Secure enclave execution introduces latency and can be costly, especially for resource-intensive AI workloads.

High Developer Learning Curve - From Phat Contracts to testnet setups, the platform requires considerable technical skill compared to traditional AI stacks.

Scalability & Adoption Risks Network Throughput Bottlenecks - TEEs and blockchains both have scalability constraints; handling large-scale AI traffic remains a future challenge.

Cold Start Problem in AI Marketplace - A decentralized AI ecosystem needs enough agents, users, and models to kickstart a vibrant marketplace—which takes time.

---

9. Reflections & Discussion

💡 Personal Reflections

One of the most surprising insights from studying Phala Network was realizing that a privacy-preserving, decentralized AI infrastructure is not just theoretical—it’s live and working.

Before this, I saw blockchain mainly as a tool for transparency and security. But combining it with TEEs showed that even sensitive AI computations can be performed privately, verifiably, and without centralized control.

This project shifted my understanding of Web3: it’s not just about token economies—it’s about reshaping how we think about data ownership, access, and digital trust in the AI era.

❓ Discussion Questions

KOR: 탈중앙화된 AI 시스템이 실제로 상용화되려면, 기술 외에 어떤 사회적·제도적 조건들이 충족되어야 할까요? ENG: What non-technical (social, legal, or economic) factors must be addressed for decentralized AI infrastructure to reach mainstream adoption?

KOR: Phala와 같은 프로젝트가 민감한 의료·금융 데이터에 적용된다면 어떤 가능성과 위험이 있을까요? ENG: If systems like Phala are applied to sensitive fields (e.g., healthcare or finance), what new opportunities—and risks—could arise?

---

10. Insight from others

After each presentation in class, we will form small groups for each case for discussion. At that time, please discuss with your group the questions posed in Section 9, and write any key points or insights from your discussion group here.

---

11. References

• https://phala.network/
• https://docs.phala.network/
• https://messari.io/report/phala-network-introduction
• https://www.intel.com/content/www/us/en/products/docs/accelerator-engines/software-guard-extensions.html
• https://www.rocketx.exchange/blog/guide-to-phala-network-buying-pha-tokens/