Privacy, Web3

Artificial intelligence systems are racing ahead, but the two questions still keep builders up at night: “Can the outputs be trusted, and is the underlying data really private?” The same tension is also relevant for blockchains, where users want verification without leaking sensitive information or relying on centralized gatekeepers.

Cryptocurrency project Polyhedra sits at the intersection of those challenges. Founded by researchers from UC Berkeley, Stanford and Tsinghua University, the team is turning zero-knowledge (ZK) proofs into production-grade tooling that can verify machine-learning results, move assets across more than 25 blockchains and safeguard user data without exposing a single secret.

Its stack already includes Expander (a high-performance ZK prover), zkPyTorch for developers and zkBridge for crosschain messaging, and it is now rolling out EXPchain, a layer-1 purpose-built for AI.

To unpack why verifiable privacy matters in AI and Web3, Cointelegraph spoke with Tiancheng Xie, Polyhedra’s co-founder and chief technology officer, ahead of the start of its zkML festival in May. A cryptographer who earned his Ph.D. at UC Berkeley, Xie leads the team’s push to turn ZK research into the “trust layer” he believes the next generation of decentralized and AI-powered applications will require.

Cointelegraph: What are the main drivers for verifiable, privacy-preserving infrastructure in AI?

Tiancheng Xie: The main drivers are security, privacy and trust powered by zero-knowledge proofs (ZKPs). As AI and blockchain technologies advance, the need to protect sensitive data while maintaining transparency and accountability grows. In AI, privacy-preserving techniques, such as ZKPs, ensure that models can be trained and predictions can be made without exposing confidential information.

In blockchain, verifiable privacy is essential to secure transactions while maintaining user anonymity. ZKPs help verify transactions without revealing transaction details, preserving privacy and security.

With these technologies, AI models and blockchain networks can operate in a trustless environment, ensuring that users and data are protected while allowing reliable and verifiable interactions.

CT: How do ZKPs hold machine learning models accountable without exposing their underlying data?

TX: ZKPs hold machine learning models accountable by verifying the correctness of computations or predictions without revealing the underlying data. In traditional machine learning, models are trained on sensitive data, and sharing the results can expose private information.

ZKPs, however, enable a model to prove that it has made a correct decision or followed a valid process without disclosing the data used to train or test it. For example, a machine learning model can prove its output is consistent with a set of rules or conditions (like a classification result) without revealing the model parameters that led to that decision.

This ensures accountability, as the model’s behavior can be independently verified, while maintaining the privacy and confidentiality of the data it was trained on, making it ideal for use in privacy-sensitive applications.

CT: What skills will tomorrow’s zero-knowledge machine learning (ZKML) engineers need, and how is Polyhedra actively nurturing that talent pool?

TX: ​Tomorrow’s zkML engineers will need a blend of specialized skills and a strong foundation in machine learning and cryptography. Key competencies include:​

  • Proficiency in PyTorch: As zkML often integrates with PyTorch, familiarity with this framework is essential.​

  • Understanding of ZKPs: A solid grasp of ZKPs is crucial for verifying computations without exposing data.​

  • Knowledge of zkML tools: Familiarity with tools like zkPyTorch, which simplify the integration of ZKPs into ML models, is beneficial.

  • Experience with ZK provers: Understanding ZK provers, such as Polyhedra’s Expander, is important for efficient proof generation.​

Polyhedra is actively nurturing this talent pool through initiatives like the Explore Expander Bootcamp and Hackathon. This six-week program, launched in collaboration with the Ethereum Foundation, Worldcoin and Google, offers hands-on training and mentorship from industry leaders.

Participants engage in deep-dive sessions on topics like ZKPs and zkML, and work on projects that apply these concepts in real-world scenarios. The bootcamp aims to equip engineers with the skills to build transformative zero-knowledge applications, fostering the next generation of zkML talent.

CT: What does it practically mean to build a “trust layer” for AI and the wider internet, and why is that mission urgent?

TX: Building a "trust layer" for AI and the wider internet means creating a system where data, transactions and computations can be verified as accurate without relying on a centralized authority.

For AI, this involves ensuring that machine learning models are transparent, accountable and operate within defined ethical and privacy standards. For the internet, it refers to establishing trust in decentralized platforms, ensuring users can interact safely and with confidence.

The urgency of this mission stems from the growing reliance on AI and decentralized technologies in critical sectors like healthcare, finance and governance. As these technologies become more integrated into daily life, the risk of misinformation, data breaches and biased algorithms increases.

A robust trust layer helps prevent exploitation and ensures that users’ privacy and rights are protected, enabling the safe and responsible adoption of AI and decentralized systems on a global scale.

CT: How do you balance the push for transparency with the right to privacy in a zero-knowledge-driven online world?

TX: Balancing transparency with privacy in a ZK-driven world requires a careful approach that ensures accountability and confidentiality. ZKPs offer a solution by allowing parties to prove something is true without revealing the underlying data. This means that individuals can maintain privacy while ensuring that systems are transparent and trustworthy.

For example, in a financial transaction, a user can prove they have enough funds for a transaction without revealing their balance or personal details. This preserves privacy while maintaining transparency that the transaction is valid.

In an online world driven by ZKPs, transparency comes from the ability to verify actions or data in a verifiable way, while privacy is preserved by only sharing what’s necessary. The key is finding ways to verify processes and outcomes while limiting exposure to sensitive information, creating a balance between openness and confidentiality.

CT: What have been Polyhedra’s toughest technical or market challenges, and which achievements confirm the strategy is working?

TX: Polyhedra has faced key challenges in both technical development and market adoption. Technically, building a scalable zkML system was tough, as early solutions were slow. To address this, Polyhedra developed Expander, a high-performance ZK prover, which significantly improves transaction speeds.

Another challenge was making zkML accessible to developers. To address this, we developed zkPyTorch, a tool that integrates with PyTorch, enabling developers to build zkML applications more easily.

In the market, Polyhedra faced the challenge of gaining trust and adoption for zkML and zkBridge, its crosschain solution. Despite this, zkBridge now facilitates millions of transactions across more than 25 blockchains, proving its value.

Key achievements that confirm Polyhedra’s strategy are the launch of the EXPchain testnet for zkML applications and strong partnerships with many leading institutions.

CT: Could you walk us through key technologies and products that shape Polyhedra’s roadmap?

TX: Polyhedra's roadmap is shaped by key technologies and products designed to enhance scalability, privacy and interoperability in blockchain and AI. These include:

  • EXPchain: A layer-1 blockchain built for AI applications, integrating zkML, the Expander proof system, zkPyTorch and zkBridge for seamless crosschain interoperability.

  • Expander: A high-performance ZKP system that enables real-time verification of complex AI computations, crucial for scalability.

  • zkPyTorch: A toolkit that makes it easier for AI developers to integrate ZKPs into machine learning models, streamlining the development of verifiable AI applications.

  • zkBridge: A trustless crosschain solution that facilitates secure data and asset transfers across multiple blockchains, addressing the challenge of trustless interoperability.

  • Proof Cloud: A cloud-based computing platform designed to provide scalable ZKP capabilities for a range of applications.

These innovations help address critical challenges in AI security, data privacy and blockchain interoperability.

CT: How will everyday interactions with AI-powered applications change in the future if Polyhedra realizes its vision of a global trust layer?

TX: If Polyhedra realizes its vision of a global trust layer, everyday interactions with AI-powered and blockchain applications will become more secure, seamless and privacy-preserving. Users will be able to interact with decentralized systems and AI models confidently, knowing that their data is protected and their actions are verifiable without exposing sensitive information.

For instance, all AI systems will provide accurate predictions and services while maintaining privacy, with verifiable accountability for their decisions. Blockchain applications will enable frictionless crosschain communication, allowing assets and data to move securely between networks.

Learn more about Polyhedra

Disclaimer. Cointelegraph does not endorse any content or product on this page. While we aim at providing you with all important information that we could obtain in this sponsored article, readers should do their own research before taking any actions related to the company and carry full responsibility for their decisions, nor can this article be considered as investment advice.