Building the future of intelligence: January 2025

Just one month into 2025, AI continues to accelerate at an exponential rate. DeepSeek's recent breakthrough in AI technology, particularly with the cost-efficient development of its R1 model, underscores the potential for AI to become more widely accessible and economically viable, especially for AI native applications. By significantly reducing the financial barriers traditionally associated with AI innovation, DeepSeek is paving the way for broader adoption across various industries, from blockchain applications to decentralized finance. This democratization of AI aligns strongly with our thesis: as AI becomes more integrated into everyday use cases, the demand for reliable verification mechanisms will only increase.

Following the surge of DeepSeek attention and usage, cloud security firm Wiz reported a major exposure in the database infrastructure, accessible from the public internet with no password required. The exposure spanned a significant volume of chat history, backend data, and sensitive information, including log streams, API Secrets, and operational details. In their blog, Wiz were clear that the rapid adoption of AI services without corresponding security is inherently risky, and as AI native systems become deeply integrated into businesses worldwide, the industry must recognize the risks of handling sensitive data and enforce security practices on par with those required for public cloud providers and major infrastructure providers.

With AI models being deployed in more contexts, ensuring the trustworthiness of AI-generated data is paramount, reinforcing the need for infrastructure to verify these systems - enhancing their reliability and enabling safer integration into diverse ecosystems. A recent article from our Chief Cryptographer explored the urgent need for accountability in-step with AI progress, and the d/acc framework’s practical steps to build trustworthy systems.

Laying out the blueprint

At the start of January we released our 2025 Roadmap, outlining a strategic path for integrating zero-knowledge proofs (ZKPs) into AI - reinforcing trust, scalability, and privacy at every layer. At the core of this vision is EXPchain, a blockchain protocol designed to handle verifiable and privacy-preserving AI operations at scale. For our first set of benchmark’s we proved Llama-3 with 8 billion parameters in around 2.7 hours on a single core CPU. Currently the team is implementing the same prover in CUDA, with expected completion at the end of February. The preliminary results are impressive, showing an improvement of 300x-2800x from the single-core benchmark.

Our roadmap leverages the Expander proof system to accelerate zero-knowledge machine learning (zkML) innovation. To date, Expander has undergone extensive testing and reviews to validate its performance, and as we prepare for broader deployment we’ve launched a $500K Bug Bounty to further harden the system against vulnerabilities.

zkPyTorch, a toolkit simplifying zkML integration into mainstream AI workflows. zkPyTorch makes it easy to turn any AI model using this compiler into ZKP-enabled AI, so that anything AI does can be resolved on-chain. With zero-knowledge proofs, we can prove AI is working responsibly and correctly without leaking any underlying data.

zkBridge, our cross-chain interoperability solution strengthening secure communication between blockchains, is also fundamental in our development of Expanded Proof of Stake (ExPoS). ExPoS is the consensus backbone of EXPChain, enabling us to unify and connect staking mechanisms across all blockchains into a cohesive network for omni-chain staking.

To help navigate our zkML-enabled L1, we’ve released EXPcenter on Telegram as our inhouse directory for all things EXPchain.

Explore Expander Bootcamp + Hackathon

On Jan 20 we launched the Explore Expander Bootcamp and Hackathon in collaboration with Ethereum Foundation, Worldcoin, Google and a panel of industry leaders. This six-week immersive program provides developers with a unique opportunity to harness Expander and drive innovation in blockchain, cryptography, and verifiable AI. A key focus is Expander’s application to the highly anticipated Ethereum Foundation proposal for the BEAM chain.

Highlights from early sessions

The bootcamp has already featured deep-dive sessions from leading experts in cryptography and blockchain infrastructure:

  • Jens Groth (Nexus) provided an introduction to zero-knowledge proofs, laying the groundwork for advanced discussions.
  • Justin Drake (Ethereum Foundation) led an in-depth workshop on native rollups, breaking down state roots, data availability, sequencing, and the execution layer of Ethereum’s world computer.
  • Ryan Cao (Worldcoin) delivered an insightful session on zkML and privacy-preserving machine learning, exploring the intersection of AI and cryptography.

The response from developers and researchers has been overwhelmingly positive:

"Hi, thanks for doing this Bootcamp-Hackathon, the workshops so far have been amazing."

Ryan Cao shared key insights on how zkML is transforming verifiable machine learning:

"What we are doing in zkML… is we are trying to show that the person running the machine learning model did so in the correct manner, with model integrity."

He also highlighted the privacy-preserving capabilities of zero-knowledge proofs:

"Because of the ZK feature of zero-knowledge proof systems, it is possible to hide certain things… Let’s say you just spent millions training the latest language model—you might not want people to know the model weights, but you still want everyone to know you trained the model correctly."

Explore Expander has now entered the Build Phase

Through this phase of the bootcamp, participants will build projects in 1 of 5 tracks: 

  • - Efficient Onchain Verifiers Using Recursive GKR Compression
  • - Account Abstraction with zkLogin (Keyless Login)
  • - Single Slot Finality: Post-Quantum Signature for BEAM Chain Consensus
  • - Building zkML with GKR Protocol
  • - Post-Quantum Private Transactions on Ethereum

The leading project in each track will receive a $50K grant. Closing talks will be given by Vitalik and myself, with all previous talks now available on the event page.

Scaling EXPchain partner integrations

January has seen a steady run of partner announcements, including:

  • Zypher Network launching the first verifiable 2048 puzzle live on EXPchain Testnet.
  • A fresh teaser from Sushi.
  • KNN3Network building smarter, secure blockchain ecosystems with a gateway to zkML-powered AI and zkBridge-enabled cross-chain data curation.
  • Openledger pushing the limits of model development with decentralized trust infrastructure for the world’s data and the power of zkML.
  • Dynamic powering seamless connectivity to EXPchain for blockchain natives and newcomers alike, enabling logins with your crypto wallet or an existing social login.
  • Cognitive Finance exploring new horizons of AI-driven agents and intent-based actions for secure, streamlined management.
  • AIdea Game building the first Verifiable AI Game Framework on EXPchain for fair play and transparent rewards powered by our zkML technology.
  • Lendvest using deterministic ZK proofs to address the state of DeFi transactions.
  • RentaNetwork shaping the future of decentralized infrastructure to redefine scalability, security, and blockchain connectivity.
  • Lemonade Social powering loyalty at the intersection of social and community.

As EXPchain moves steadily towards Mainnet, we’ve released Devnet to continue testing code in a sandbox environment. 

The Polyhedra Show: discussions on the Future of Intelligence 

Since December, The Polyhedra Show has been bringing together researchers, engineers, and ecosystem pioneers to explore the technologies shaping the future of Zero-Knowledge Machine Learning, blockchain trust, and AI security. Airing biweekly on Tuesdays and Thursdays, each fireside unpacks the latest breakthroughs in verifiable computing, privacy-preserving AI, and decentralized infrastructure:

Polyhedra 2025 Roadmap - Jan 7th

I’m joined by our CMO Samuel Pearton to discuss Polyhedra’s 2025 roadmap, focusing on EXPchain and innovations in zkML, AI trust, blockchain scalability, and key advancements with zkPyTorch, enabling verifiable AI models without exposing private data, Proof of Intelligence (PoI) to ensure AI model integrity, and a trustless AI Data Marketplace. 

Explore Expander: Tracks - Jan 9th

Our Chief Cryptographer Zhenfei Zhang joins CSO Eric Vreeland to share insights into the Explore Expander boot camp focus areas - including single-slot finality, post-quantum security, and zkML - along with the boot camp's structure, importance in accelerating Ethereum’s development, and how participants can get involved.

Verifiable Gaming with Zypher Network - Jan 14th

Zypher Network COO Maggie Wang joins our CSO Eric Vreeland to discuss the collaboration between Polyhedra and Zypher Network, the role of ZK technology in ensuring fairness and transparency in gaming, and the launch of z248, a verifiable version of the classic 2048 puzzle game.

Introducing zkML - Jan 16th

Our Chief Scientist Jiaheng Zhang and Strategy and Operations Lead Morgan Takach explore zero-knowledge machine learning (zkML), a novel approach combining AI with cryptographic proofs, discussing how Jiaheng’s research on zkML enhances AI verifiability and privacy. 

Verifiable AI Agents - Jan 21st

Our CMO Samuel Pearton joins me one more to focus on AI agents and the need for verifiability, privacy, and ethical decision-making, discussing how Polyhedra is creating an infrastructure layer to ensure AI systems remain trustworthy, using ZK proofs to safeguard user data.

Building Trust in AI Data with OpenLedger - Jan 23rd

OpenLedger co-founder Ram Kumar joins our CSO Eric Vreeland to discuss Polyhedra and OpenLedger’s collaboration to build a decentralized trust infrastructure for AI data, and "proof of attribution," ensuring fair compensation for data contributors and enhancing the trustworthiness of AI models.

Evening Edition: Community Q&A - Jan 28

This Q&A episode with our Strategy and Operations Lead Morgan Takach and Head of Engagement Anthony Cargill addresses community questions about AI, blockchain, and quantum computing, covering the $500 billion Stargate project, ethical AI, and the impact of ZK proofs on trust in AI.

Building Seamless UX with Dynamic - Jan 30

Dynamic co-founder Itai Turbahn joins our CSO Eric Vreeland to discuss Polyhedra’s partnership with Dynamic to improve blockchain user experience by integrating decentralized security with intuitive wallet interactions and removing the friction between usability and security, making blockchain applications more accessible while preserving decentralization and trust.

January has set a fast pace - EXPchain Devnet is live, zkML benchmarks are breaking records, and ecosystem adoption is growing. With research, development, and partnerships accelerating, 2025 is shaping up to be a defining year for verifiable AI and blockchain scalability.

-Tiancheng Xie, Polyhedra co-founder 

Expander zkML EXPchain