Here we report on the progress of the leading builders in the ZK/AI ecosystem, documenting recent significant releases, technical breakthroughs and general updates.
Featuring: @0xHolonym, @inference_labs, @lagrangedev, @PolyhedraZK, @spaceandtime, @zk_agi, & @zkLink_Official
@0xHolonym published an article covering its ongoing 10-day residency at Edge City Patagonia in Argentina, where 25 builders, artists, and researchers are collaborating to create technologies aligned with human values: https://human.tech/blog/building-the-future-in-the-andes-the-human-tech-residency-at-edge-city-patagonia
The program highlights AI and machine learning for social benefit - systems that remain transparent, accountable, and centered on human agency.
Guided by the d/acc philosophy and the Covenant of Human-Aligned Technologies, participants are prototyping privacy, governance, and ethical AI tools, supported by up to $10,000 in HUMN grants.
JSTprove
@inference_labs has introduced JSTprove, a new open-source toolkit that lets anyone verify AI model outputs using zkML: https://x.com/inference_labs/status/1983925569495433656
The system simplifies complex cryptographic processes, allowing developers to confirm that AI results match the original model’s logic without exposing private data.
Built on @PolyhedraZK’s Expander backend, JSTprove supports ONNX models and includes an easy command-line interface. Future updates will add support for larger architectures and GPU acceleration.
Full details: https://arxiv.org/abs/2510.21024
Publications
The project has published three articles exploring how cryptographic methods can bring trust, transparency, and accountability to AI systems:
• In the first article, the focus is on the real AI crisis - the lack of verifiability. It explains that the main risk isn’t hallucinations but the inability to prove what AI models actually do, emphasizing Proof of Inference as the key to building auditable and trustworthy AI: https://x.com/inference_labs/status/1975964815819505737
• The second article highlights the fragility of AI content watermarking, showing how it can be easily removed or forged. It argues that only cryptographic verification can ensure authentic, tamper-proof content and restore trust in AI-generated media: https://x.com/inference_labs/status/1980695118945325436
• In the last article, the discussion centers on ZKPs and their role in enabling privacy-preserving and verifiable AI computations. It also outlines their current limitations in scalability and usability while stressing their importance for secure and decentralized AI systems: https://x.com/inference_labs/status/1986116116104814686
DeepProve
Lagrange has integrated its DeepProve system into Anduril’s Lattice SDK to demonstrate verifiable AI decision-making within autonomous reconnaissance workflows: https://lagrange.dev/blog/deepprove-anduril-lattice-sdk
The project combines trained machine learning models with ZKPs, allowing every tactical action - such as proximity detection, response classification, and movement calculation - to be cryptographically verified without revealing sensitive data.
This prototype highlights how defense systems can pair autonomy with real-time verification for transparent and accountable AI operations.
Events
1/ Lagrange has extended its DeepProve system to verify @Google’s new Gemma 3 AI model, announced at the Verifying Intelligence Event in Singapore, hosted by @HouseofZK and @boundless_xyz in partnership with @googlecloud: https://lagrange.dev/blog/proving-gemma-3
Gemma 3’s compact architecture improves speed and accuracy while running efficiently on devices. DeepProve’s integration enables cryptographic verification of Gemma 3’s outputs, allowing advanced models to deliver provable, trustworthy results across real-world applications.
Keynote by the Founder @Ismael_H_R from the event: https://x.com/HouseofZK/status/1978025917407465942
2/ He also joined @alicelingl of HoZK, @DaLiberman of @gonka_ai, @DMSKwak of @LazAINetwork, and @robviglione of @ZKVProtocol to discuss how cryptography can make AI verifiable and trustworthy.
Full panel: https://x.com/HouseofZK/status/1979104354196947195
Partnerships
Lagrange has announced multiple new partnerships, including:
• @arbitrum: Integrated DeepProve into Arbitrum to support onchain AI agents participating in DeFi use cases. The collaboration powers a new generation of autonomous agents capable of managing portfolios and executing trades with verifiable logic: https://x.com/lagrangedev/status/1973463672870805791
• @usetoku: Teamed up to bring verifiable AI to global payroll systems. DeepProve adds cryptographic proof to AI-driven wage and tax calculations, enabling autonomous, auditable, and privacy-preserving payment workflows using stablecoins: https://x.com/lagrangedev/status/1976652509021360201
• @MAIN_AI_DEX: Partnered with Lagrange to integrate DeepProve, allowing AI-driven DEX trades to be cryptographically verified. This ensures every trading action by an agent is transparent, auditable, and proven without exposing sensitive data: https://x.com/lagrangedev/status/1984327424142704998
• @khalani_network: Integrated Lagrange’s zkML to make intent execution verifiable across solver networks. Developers can now cryptographically prove cross-chain coordination outcomes, bringing audit-grade trust to intent-driven DeFi flows: https://x.com/lagrangedev/status/1983584303780151692
• @GrovioAI: Teamed up with Lagrange to verify AI-generated marketing decisions. DeepProve enables Grovio to prove that AI outputs are accurate, privacy-preserving, and compliant - without slowing down creative workflows: https://x.com/lagrangedev/status/1983217987914985523
• @MerkleScience: Joined forces with Lagrange to bring zero-knowledge verification to blockchain risk analysis. AI-generated alerts and risk scores now come with cryptographic proofs, adding verifiable trust to compliance and fraud detection: https://x.com/lagrangedev/status/1981410242907828679
• @Talus_Labs: Integrated DeepProve into Talus’ agent infrastructure to verify AI decisions on-chain. This enables trusted agent activity in gaming, betting markets, and autonomous workflows - particularly within the Sui/Move ecosystem: https://x.com/lagrangedev/status/1981050399298601190
@PolyhedraZK has applied its zkPyTorch framework to the Gemma-3 model, enabling verifiable AI inference without revealing model data or parameters: https://blog.polyhedra.network/data-protective-proofs-for-next-generation-ai-proving-gemma3-with-zkpytorch/
By converting Gemma-3 into ZK-friendly circuits, zkPyTorch allows users to confirm that outputs are genuine while protecting intellectual property.
The system employs quantization, preprocessing, and hierarchical optimization to ensure efficient proof generation. This integration highlights how Gemma-3 can operate securely in sensitive fields where data protection and verifiable computation are essential.
Space and Time v2
@spaceandtime has launched v2 of its mainnet, enabling institutions to securely connect and verify offchain financial data for tokenized assets: https://spaceandtime.io/blog/space-and-time-announces-mainnet-v2-with-custom-offchain-data-tables-for-institutional-customers
The upgrade addresses risks tied to using centralized systems for linking asset data to blockchains, which can lead to mispricing or faulty settlements.
With v2, banks, asset managers, and enterprises can ensure data integrity for tokenization and stablecoin projects. Major institutions, including @Microsoft, have already adopted Space and Time’s data solutions.
Space and Time has announced two new integrations expanding the reach of its verifiable data infrastructure across financial and educational ecosystems.
Integrations
• @StellarOrg: Integrated Stellar into Space and Time’s blockchain indexing layer, allowing developers to access verifiable onchain and cross-chain data for building advanced financial applications. The collaboration enables smart contracts on Stellar to reference historical data and external sources securely, supporting use cases like payments, asset issuance, and token flows: https://spaceandtime.io/blog/space-and-time-enables-new-sophisticated-financial-apps-for-the-stellar-ecosystem
• Salib Suci Foundation: Partnered with Indomobil Group and Space and Time to store and verify English fluency test credentials for 14,000 K-12 students across 70 schools in Indonesia. Students use SXT, the native token of Space and Time, for course enrollment, while course results are recorded on-chain for universal verifiability. The initiative aims to expand access to trusted education credentials across the country: https://spaceandtime.io/blog/salib-suci-foundation-to-verify-course-results-for-14-000-students-across-70-schools-with-space-and-time
Vision
@zk_agi published an article explaining its vision of building private artificial intelligence through a privacy-focused protocol that ensures confidentiality, composability, and trusted infrastructure: https://x.com/zk_agi/status/1978055404866371719
The project aims to create an ecosystem where developers can build AI agents and applications that think and learn securely, protecting both innovation and user data.
Through its DAO governance and open collaboration, ZkAGI seeks to make privacy a foundation for intelligent systems and long-term community-driven growth.
Partnerships
ZkAGI has partnered with @DelphinusLab to explore integrating ZKWASM technology into its privacy-preserving AI network: https://x.com/DelphinusLab/status/1977585661432783311
The collaboration focuses on enabling verifiable and confidential AI inference, ensuring every model action can be proven without revealing sensitive data. Combining ZkAGI’s decentralized GPU network with Delphinus Lab’s verifiable compute layer, the partnership advances accountable AI infrastructure.
