How Zero-Knowledge Proofs Are Revolutionizing AI and DeFi: Insights from Brevis CEO

Zero-Knowledge (ZK) proofs are emerging as a transformative technology, offering unprecedented solutions for challenges in artificial intelligence (AI) and decentralized finance (DeFi). Michael Dong, CEO of ZK proof startup Brevis, recently highlighted the pivotal role of ZK proofs in enabling “verifiable computing” during the “AI on AIR” podcast, hosted by Kite AI. Dong likened ZK proofs to “painkillers” essential for addressing AI trust issues and bolstering DeFi ecosystems, contrasting them with “vitamins” that merely contribute to scaling rollups.

Transforming Trust in AI and DeFi Through ZK Proofs

Brevis, a pioneer in ZK proof technology, focuses on verifying off-chain computations on-chain, addressing critical trust deficits caused by modern AI's “black box” nature. As AI becomes integral to sectors like healthcare, education, and finance, accountability is paramount. According to Dong, ZK proofs offer the only viable solution, enabling “verifiable inference”—a transparent mechanism to validate AI outputs and eradicate trust issues.

Dong expressed concern about the lack of transparency inherent in many AI systems, where outputs are generated without assurance of integrity. ZK proofs serve as a bridge to secure trust in these outputs, using cryptographic methods to validate authenticity while preserving data privacy.

The Evolution of ZK Proof Technology: Economic Efficiency and Scalability

ZK proof technology has undergone significant advancements in recent years, particularly in reducing computational costs. Dong explained that generating a ZK proof previously required costs nearly one million times higher than the original computation. Thanks to breakthroughs in cryptographic research and hardware acceleration, specifically advancements in GPUs, this cost has been reduced to 1,000 times.

The resulting efficiency highlights the superior performance of ZK proofs compared to traditional blockchain setups. For example, Ethereum relies on over 800,000 nodes executing repetitive computations, making the process highly inefficient. ZK proofs, on the other hand, enable a single node to generate proofs, which the rest of the blockchain network can verify at minimal cost. This optimization could drastically enhance scalability.

Brevis has improved ZK proof generation capabilities with its proprietary tool, Pico Prism, which generates Ethereum block proofs in just six to 15 seconds. With enhanced GPU resources, this technology could theoretically scale Ethereum block sizes by 100x and lower gas fees by the same margin. Beyond Ethereum, Dong noted that blockchains like Solana, which rely on centralized nodes for efficiency, could benefit from ZK proofs by reducing validator dependency and boosting decentralization. Additionally, ZK proofs can enable lightweight devices like smartphones to validate blockchain states, eliminating reliance on less secure remote nodes.

Building User Trust in AI Systems With “Verifiable Inference”

Incorporating ZK proofs into AI systems introduces a paradigm shift in securing user trust. Dong explained how ZK proofs allow users to verify AI-generated outputs, ensuring recommendations or decisions are credible. For example, take a dubious medical suggestion from an AI doctor, such as “consume six bananas a day.” Users would have the ability to request ZK proof-backed verification, confirming whether the advice derives from an unaltered, trusted AI model.

This process relies on pre-verified AI systems registered under a ZK “verifier key.” When questioned, the AI responds with a ZK proof ensuring the integrity of its output, enabling transparency combined with cryptographic security. As Dong emphasized, ZK proofs deliver 100% confidence that AI outputs stem from their original trustworthy models.

Currently, executing machine learning (ML) models directly within zero-knowledge Ethereum virtual machines (ZK-EVMs) is operationally unfeasible due to resource constraints. Brevis addresses this limitation through a modular architecture that integrates ZK-EVM as an overarching layer while external co-processors handle specialized computations. For example, verifying 30 days of Uniswap trading data could involve parallel processing of individual transaction proofs through a ZK data co-processor, while a simplified ZK-EVM program aggregates the results. Dong suggested that future ZK-ML operations would likely adopt similar architectures, using external processors for seamless integration between machine learning and zero-knowledge systems.

Stablecoins: The Ideal Payment Method for AI Agents

Expanding on the intersection of blockchain and AI, Scott Shi, co-founder of Kite AI, discussed the evolving landscape of financial transactions for autonomous AI agents. Shi emphasized the inadequacy of traditional payment methods, such as credit cards, for AI agents that require frequent, micro-sized transactions. Instead, blockchain-based stablecoins are the most suitable “native currency” for AI agents due to their programmability, low fees, and operational maturity.

To facilitate these transactions, Kite AI is developing specialized tools such as “Agent Passports,” “Account Abstraction” solutions, and a dedicated “Stablecoin Payment Lane.” These innovations aim to provide isolated block space for agent interactions, ensuring predictable and affordable gas fees. Stablecoins are not only a logistical solution but also hold the potential to redefine how AI gathers and transfers value in blockchain-powered ecosystems.

Collaborative Potential Between Brevis and Kite AI

Highlighting synergies between the two companies, Dong revealed how Brevis’ ZK technology could enable trust-centric AI agent reputation systems by proving their historical behavior on-chain. Kite AI, exploring these possibilities, expressed enthusiasm for partnerships with Brevis. According to Shi, “There’s immense potential for collaboration, and groundbreaking outcomes may arrive soon.”

Key Takeaways: Scaling Transparency Across AI and Blockchain

The discussion underscores ZK proofs as an indispensable innovation for enhancing transparency and scalability in both AI and blockchain ecosystems. From enabling “verifiable inference” in AI systems to optimizing blockchain efficiency and decentralization, ZK proofs are positioned to address fundamental trust issues and open new opportunities for growth across industries. As Brevis and Kite AI continue exploring collaborative projects, the integration of ZK technology could reshape user confidence and operational models in the AI and DeFi sectors.