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: @lagrangedev, @NexusLabs, @SpaceandTimeDB, @zk_agi, @gizatechxyz, @PolyhedraZK, @cysic_xyz, @SindriLabs, @alignedlayer, & @ZircuitL2
Research
@lagrangedev has received U.S. patent No. 12,368,596 for Reckle Trees, a cryptographic structure enabling updatable batch proofs that significantly outperform traditional Merkle proofs: https://x.com/lagrangedev/status/1948101548053291176
Integrated into Lagrange’s coprocessor, Reckle Trees allow incremental updates without full recomputation, boosting performance up to 200× in dynamic workloads. The technology has drawn academic recognition and is being benchmarked by other researchers.
Publications
@CoinDesk published an opinion piece by @Ismael_H_R, CEO of Lagrange, explaining how zero-knowledge machine learning could counter rising deepfake threats: https://x.com/CoinDesk/status/1945996947476881841
In the article, Ismael cites a $46 million Hong Kong scam and billions lost globally as proof of the situational urgency.
zkML allows AI moderation decisions to be verified without revealing sensitive data, enabling faster, portable, and trustworthy checks across platforms while preserving privacy and reducing computational waste.
Media
In a recent episode of @HouseofZK Radio, recorded in person at The ZK/AI Summit during @token2049 Dubai, Ismael joined @alicelingl to discuss DeepProve - Lagrange’s high-performance zkML framework, and its role in ensuring AI outputs are provably correct: https://x.com/HouseofZK/status/1952729087924994442
They explored the importance of verifiable AI models, securing every ML pipeline stage with ZKPs, and the technological convergence of ZK, AI, and crypto.
Another recent episode featured Franklin Delehelle, Research Engineer at Lagrange, who talked about the company’s shift from its zkSQL co-processor to a modular, proof-system-agnostic prover network supporting zkRollups: https://x.com/HouseofZK/status/1947608401669710196
Franklin explained its first-principles design, modular advantages, incentive alignment via DARF auctions, cross-chain queries, complexity abstraction for developers, and future directions inspired by hardware and cloud computing.
Partnerships
Lagrange have announced several new partnerships and integrations, including:
@buildonbase: Integrated DeepProve to enable AI agents to prove outputs onchain with privacy, cryptographic verification, and up to 1000× faster performance: https://x.com/lagrangedev/status/1943309251457560836
@SentientAGI: Powered Sentient Chat with DeepProve to verify AI responses onchain while keeping inputs and model weights private: https://x.com/lagrangedev/status/1945116589554077893
@OpenledgerHQ: Deployed DeepProve to make trading predictions and smart contract optimizations verifiable onchain without exposing sensitive data: https://x.com/lagrangedev/status/1946195681980346785
@Wach_AI: Enabled adversarial simulations of autonomous agents to be cryptographically proven and verified onchain: https://x.com/lagrangedev/status/1948384440457019791
@hetu_protocol: Used DeepProve to prove datasets, model updates, and research contributions, ensuring provenance and transparency: https://x.com/lagrangedev/status/1953159146355999224
Events
@danielmarinq, CEO of @NexusLabs, and @alexanderfowler, Chief Strategy Officer, spoke at BASS 2025 and SBC’s Decentralized AI Summit:
https://blog.nexus.xyz/verifiable-ai-why-trust-must-be-proven-not-promised/
Daniel highlighted that current blockchains are ill-suited for AI agents, outlining Nexus’s purpose-built design with native language support and optimized performance.
Alexander emphasized cryptographic proofs as essential for trust, warning that weak regulation and poor accountability threaten transparency in autonomous systems.
@0xAlecJames, Business strategy manager at Nexus, also featured at an AI Collective talk, where he warned that AI development is advancing far faster than ethics and oversight: https://blog.nexus.xyz/ai-collective-talk-recap-with-alec-james-recap/
Citing 2015 data showing nearly 1,900 times more general AI research papers than on AI ethics, he outlined Nexus’s cryptographic verifiability tools, including a zkVM, to ensure trustworthy AI in healthcare, finance, and media while protecting privacy and maintaining accountability.
Partnerships
Nexus announced several partnerships to expand its verifiable AI and blockchain infrastructure, driving innovation, liquidity, and adoption across its ecosystem. These include:
@onyxdotbond: Collaborated to bring tokenization and AI solutions for SMEs, enabling revenue-generating businesses to tokenize assets, access alternative financing, and leverage verifiable AI tools for efficiency, customer engagement, and liquidity: https://blog.nexus.xyz/tokenization-and-ai-as-tools-for-smes-a-vision-of-nexus-and-onyx-partnership/
@Vishwa_xyz: Partnered to integrate the Nexus zkVM into Vishwa’s ZK network for trust-minimized credit systems, deploy credit rails on Nexus, and expand liquidity while researching credit solutions for autonomous agents: https://blog.nexus.xyz/vishwa-partnership/
@Repponetwork: Teamed up to scale Verifiable AI by making every step of Reppo’s AI alignment process auditable through ZKPs, and powering collaborative workflows and agentic systems with the Nexus zkVM: https://blog.nexus.xyz/partnering-with-reppo-ai-to-scale-verifiable-ai/
@Hive_Intel: Worked together to turn natural language into structured, verifiable blockchain data, enabling agents to reason across 60+ networks and developers to build AI systems on trusted, cryptographically sound information: https://blog.nexus.xyz/hive-intelligence-partnership/
@hetu_protocol: Partnered to bring verifiability and attribution to AI inference by combining Hetu’s adaptive consensus with the Nexus zkVM, enabling transparent scientific collaboration and composable trust in AI systems: https://blog.nexus.xyz/hetu-partnership/
Tech
@SpaceandTimeDB has released Blitzar, an open-source GPU acceleration framework designed to speed up ZKP generation: https://x.com/SpaceandTimeDB/status/1948033801680204252
Originally built to make its Proof of SQL system process a million rows in under a second, Blitzar efficiently maps ZK computations to GPUs, which outperform CPUs in parallel math. It supports MSM and IPA operations, and is available for any ZK protocol needing faster proofs.
Following the release, Space and Time integrated the framework into Microsoft Nova’s HyperKZG commitment scheme: https://x.com/SpaceandTimeDB/status/1950557259857305644
This integration accelerates onchain verification for ZK protocols, enabling GPU-based recursive proving and optimized hardware use without altering proof systems. The generalized backend boosts efficiency for rollups, coprocessors, and EVM-compatible applications, supporting the broader goal of scaling verifiable compute.
Integrations
S+T has partnered with @theblessnetwork to create AI agents that can verify every decision they make: https://x.com/SpaceandTimeDB/status/1945990957104574729
The collaboration combines Space and Time’s blockchain-based data engine with Bless Network’s global compute layer, aiming to remove dependence on unverified data and enhance transparency and trust in AI-driven processes.
The project also partnered with @Calderaxyz to integrate its indexed data and sub-second ZK coprocessor into the @zksync Elastic Chain ecosystem: https://x.com/SpaceandTimeDB/status/1947420763356803111
The partnership aims to support builders in deploying scalable, data-driven applications and protocols, using verifiable data through ZK-proven SQL from the start, enabling efficient development on the Elastic Chain.
Monthly Report
@zk_agi published a monthly report for July: https://medium.com/zkagi/zkagi-monthly-progress-report-july-2025-a0c50e4b2786
In summary:
Integrated @HyperliquidX API for automated hourly trades.
Enhanced risk controls with dynamic leverage and profit/loss thresholds.
Reduced backend latency by 30% for faster execution.
Achieved ~70% Bitcoin prediction accuracy and 0.25%-1% daily profits in tests.
Expanded partnerships to five and launched two revenue streams.
Strengthened community engagement with funding rounds and ambassador rewards.
Partnerships
The project also announced a partnership with @tenprotocol to deploy encrypted AI agents on public blockchains, protecting data, strategies, and transactions from exposure: https://x.com/zk_agi/status/1952494228711567600
Using TEN’s trusted execution environments, the project enables secure model hosting, encrypted inferencing, and autonomous execution. Current applications include Deltacore for trading and Zkare for genomics and healthcare, preserving competitive advantages through encryption.
@gizatechxyz has launched ‘Arma’, a personal agent already integrated into the @baseapp wallet: https://x.com/gizatechxyz/status/1945627586131902535
The tool autonomously manages, allocates, and optimizes assets, including stablecoin-based yield strategies, around the clock. Users can activate it via in-app chat without dashboards or toggling, enabling seamless, risk-adjusted capital growth directly within their daily wallet experience.
Giza is also now more easily accessible to tens of millions of @coinbase app users. The integration allows easier access to Giza’s autonomous finance tools, including ARMA’s 24/7 yield optimization and its expanding agent network on the onchain @base ecosystem: https://x.com/gizatechxyz/status/1953856357125312568
The project has also integrated its Autonomous Financial Intelligence (AFI) agent into @BinanceWallet, enabling active management of stablecoins: https://x.com/gizatechxyz/status/1950576061386989984
Once deposited, the ARMA agent handles yield optimization, smart rebalancing, and risk control around the clock. Users can maximize returns without manual intervention, keeping funds optimized across DeFi protocols within Binance’s secure environment.
@PolyhedraZK released an article revealing the launch of @EggDotParty, an AI-powered crypto platform aiming to unify and automate creators’ fragmented digital workflows: https://blog.polyhedra.network/the-future-of-ai-plus-and-social-liquidity-why-we-are-building-egg/
EGG offers multi-modal input, integrated AI tools, publishing, monetization, and an agent marketplace in one workspace.
Built in public for community feedback, EGG’s roadmap includes enhanced plug-ins, on-chain agent verification, and streamlined creation of storefronts, apps, and branded experiences to simplify and scale creative businesses.
Tech
@cysic_xyz has launched its Agent-to-Agent (A2A) Protocol, extending Google’s A2A spec with native crypto payments for autonomous agents: https://x.com/cysic_xyz/status/1953841175921627423
The protocol enables agents to charge, verify, and receive instant payments in real-time. Through a payment-first SDK, agents can evaluate, price, and complete tasks autonomously while ensuring transparent, verifiable compensation, streamlining coordination in the machine economy.
Publications
Cysic also shared an article explaining how its GPU-accelerated platform makes verifiable AI practical by generating cryptographic proofs of model outputs in real time: https://medium.com/@cysic/verifiable-ai-cysics-gpu-accelerated-leap-in-trusted-machine-learning-a147a0466ee2
Using ZKPs, optimized GPU arithmetic, and massive parallelism, it speeds up verification for large models like CNNs by up to 10x.
The system integrates seamlessly with PyTorch and TensorFlow, with plans for broader model support, public demos, and an open-source release.
@SindriLabs published an article exploring data responsible AI: https://sindri.app/blog/2025/07/22/data-responsible-ai/
It argues how the most effective AI operates close to a user’s real data - enabling context-aware interactions that feel natural and precise - but this raises concerns over privacy and trust.
The piece explores a “trust-context flywheel”, where better results encourage sharing more data, and stresses that truly data-responsible AI must protect sensitive information while leveraging deep contextual insights to remain competitive.
@alignedlayer shared an article explaining how ZKPs, combined with @ethereum, can address AI’s growing trust gap: https://blog.alignedlayer.com/the-era-of-ai-needs-ethereum-and-zk/
As AI takes on evermore critical decisions, proof of execution and tamper-resistance become essential. ZK verifies computations; Ethereum anchors them securely.
Aligned offers tools for model verification, proving capacity, and economic agent infrastructure, enabling AI systems and humans to coordinate, transact, and prove actions without exposing sensitive data.
@ZircuitL2 has launched a new AI-powered trading engine branded as "Hyperliquid for AI Trading", designed for fast, cross-chain execution on EVM and @solana: https://zircuit.com/blog/zircuit-ai-trading
The engine scans on-chain and off-chain data for crypto signals, automates trades, and offers one-click strategies. It claims to provide secure, passive yield through Deposit Vaults and active trading with reliable, battle-tested protection.
