MetaArena is a trusted execution infrastructure designed for AI Agents and complex interactive systems. It is built to accelerate AI adoption across domains including gaming and finance, ensuring that intelligent behaviors are verifiable, auditable, and scalable through zero-knowledge verification — delivering truly reliable AI experiences for blockchain games.

As one of the few infrastructure projects focusing on trusted AI execution in blockchain gaming, MetaArena has recently gained significant market attention and successfully completed a new round of strategic funding. This round involves multiple renowned institutions, including IBC Group, Central Research, SEI Foundation (SEI Network), Sky Wee (Sky Ventures), Stratified Capital, Pacific Meta, LC Academy, Axia8, CIW, LucidBlue, IceTea Labs, ODIG, and A1E Omega.

This funding round not only reflects strong capital market recognition of MetaArena’s vision but also further validates its potential value and strategic position in the upcoming wave of on-chain intelligent interaction upgrades.

MetaArena: Making AI Agents More Trustworthy

MetaArena is a trusted execution infrastructure centered on zero-knowledge proof (ZKP) mechanisms, designed to provide verifiable computing services for AI gaming and intelligent interaction scenarios that require trustable execution.

MetaArena consists of an off-chain computing network composed of distributed nodes and an on-chain verification engine deployed across multiple chains. When a trusted execution task arises, MetaArena dispatches AI behavior requests to the off-chain nodes for execution, generating a zero-knowledge proof (ZKP) that is subsequently verified on-chain. This mechanism ensures that input data, reasoning processes, and execution results are authentic, trustworthy, and tamper-proof. MetaArena has been validated in the Web3 gaming field and enables AI Agent-driven blockchain games to run efficiently, securely, and auditable without relying on centralized servers.

In its recent upgrade, MetaArena introduced a new trusted execution stack. Through two core capabilities — zkTrace and zkAction — it verifies the consistency of prompt inputs (Proof of Prompt) and the trustworthiness of reasoning behaviors (Proof of Inference), ensuring the authenticity and confidentiality of AI Agents’ prompts and inference paths in a provable manner.

Notably, while many existing solutions attempt to provide trustworthy environments for AI Agents, MetaArena is among the few that achieve trusted execution purely through zero-knowledge cryptography without requiring specialized hardware.

zkTrace: Proof of Prompt Input Trustworthiness

In traditional AI Agent models, a core problem has remained unresolved: how to ensure the trustworthiness of prompts?

This includes, but is not limited to:

MetaArena provides verifiable and trusted execution capabilities for prompts at the computation layer through the zkTrace module, ensuring that prompts maintain correctness, consistency, and privacy throughout their lifecycle; there is no need to expose the original content externally. This is a key foundational component for building trustless AI Agents and decentralized application logic.

zkTrace is offered in a developer-friendly SDK form. Its underlying design relies on strong cryptographic mechanisms and ZK primitives, including Pedersen commitments, Poseidon, and zkSNARKs (Plonk), and works closely with the System Prompt initialization process.

During system initialization, the prompt is input into the off-chain computing network to generate a cryptographic commitment and construct a corresponding ZKP. These ZKPs can be referenced by any user or third-party verifier, and by comparing them with the on-chain prompt commitment, the authenticity and untampered state of the prompt can be confirmed. If the prompt used in execution does not match the audited commitment, verification immediately fails, ensuring transparency and trusted execution without exposing plaintext.

In practice, AI Agent developers or AI prompt application developers can use zkTrace to create and define a System Prompt, ensuring the model executes tasks strictly according to established policies and constraints. Once the System Prompt is initialized and loaded into the model, zkTrace automatically generates the commitment and proof files, submitting them to the on-chain verification engine. This process records the full lifecycle of the prompt from input to usage, ensuring proofs are traceable and tamper-proof.

For end-users interacting with AI Agents, they can always access the prompt commitments and proofs corresponding to the currently executing model and verify the authenticity of prompt usage:

Does it still match the developer’s intended settings?Has it been replaced or injected with malicious content during execution?

zkTrace ensures that prompt trust no longer relies on centralized custody or a single service provider endorsement but is instead established through cryptographic proofs, creating a verifiable, auditable, and non-repudiable foundation of trust for system inputs.

zkTrace Interaction Example

zkTrace establishes a reliable interaction mechanism among AI Agents, off-chain computing networks, DApps, and smart contracts, ensuring the integrity and consistency of prompts and providing verifiable trust guarantees for AI model behaviors.

When AI Agent developers define and submit a System Prompt via zkTrace, the prompt is encrypted off-chain and a commitment is generated, while the agent is initialized and bound to the corresponding verification circuit, giving the prompt an immutable attribute throughout the system. The AI Agent also registers the necessary verification keys with the MetaArena off-chain computing network for subsequent verification calls.

When a DApp initiates a message or interaction request, the AI Agent reads the request and delegates execution tasks to off-chain computing nodes. During execution, the usage and logic of the prompt are verified via zero-knowledge proofs, and the behavior path is recorded to generate verifiable proof files. The proof results are then returned to the smart contract or DApp to confirm at the contract level that the action strictly originates from the committed prompt.

MetaArena’s on-chain verification engine is responsible for matching ZKPs with commitments to ensure consistency between input content and executed behavior. If a prompt is replaced or execution policies deviate, verification fails immediately, effectively curbing potential abnormal action chains. This mechanism ensures that the AI Agent’s execution aligns perfectly with the initial settings and provides a transparent, auditable, and trusted foundation.

By collaborating with smart contracts and other on-chain objects, MetaArena makes AI Agent execution publicly verifiable, providing high security and structured trust for various Web3 use cases.

From a capability perspective, zkTrace enables AI Agents to have:

Based on zkTrace’s trusted input advantages, capabilities can naturally extend to Proof of Inference (implemented via zkAction) to verify the trustworthiness of AI Agent reasoning paths and results, ensuring outputs are strictly derived from legitimate input reasoning.

Overall, zkTrace is especially suitable for critical mission scenarios, such as financial-sensitive, tightly constrained, or high-compliance decision-making tasks, providing a highly secure and transparent operational foundation for the next generation of trustless AI Agents.

AI Agent Game Engine Trusted Framework

MetaArena has been first to implement in the blockchain gaming field, launching the AI Game Engine component, which constrains and audits in-game agent operations through zero-knowledge proof mechanisms. Game agents can participate in on-chain battles via smart contracts, with behaviors verified through zkTrace / zkAction to ensure fairness, authenticity, and traceability.

Within this game engine system, developers can continue using native engines like Unity, Cocos Creator, and Unreal without changing existing workflows to migrate games to a trusted on-chain environment. Through SDK interfaces, developers can access MetaArena’s decentralized state layer, managing critical on-chain states including player inputs, state changes, and turn switching, and verify them in real-time via zero-knowledge proofs.

All generated content and task feedback can be processed by multiple AI Agents (e.g., content generation agents, battle agents, testing agents), enabling automated verification and dynamic game experience optimization.

All data generated during gameplay — including command inputs, state transitions, behavior logs, and content generation results — is transmitted to MetaArena’s off-chain computing network for processing and integrated into a verifiable proof structure via ZK Game SDK. Using ZK circuits (e.g., ZK Shuffle, Action Legitimacy Circuit), randomness, fairness, and rule consistency are ensured. The on-chain verification engine publicly confirms each behavior’s authenticity through zero-knowledge verification, ensuring the game execution process is tamper-proof and fully transparent.

At the computation and storage layer, MetaArena combines resource optimization components to provide high-performance support for multi-agent scenarios (AIGC, QA testing agents, data insight agents, etc.), ensuring execution efficiency and response stability under high-throughput interactions.

Ultimately, this infrastructure not only provides developers with efficient computing resources but also, through decentralized verification + intelligent behavior auditing, ensures that every game operation is verifiable, auditable, and accountable, establishing a fair and trusted on-chain AI gaming ecosystem and effectively preventing cheating, tampering, and opaque execution.

Superior Security

In building trusted AI Agents, TEE solutions are widely adopted for their hardware-isolated environments, providing some level of data privacy protection and verifiable execution. While TEE is a validated mainstream privacy solution used across domains, it has limitations for trusted AI Agent construction.

In fact, TEE solutions often rely on trusted environments and key management services provided by hardware vendors like Intel SGX and ARM TrustZone. This centralized trust mechanism makes system security highly dependent on specific vendors, introducing centralization risks. Intel SGX has previously been exposed with multiple vulnerabilities, directly threatening trust foundations. Additionally, although TEE provides isolated runtime environments, its data privacy protection remains limited. For instance, data may be intercepted during transmission to the TEE environment, and external attackers could access sensitive information via interaction interfaces. TEE design is also primarily for predefined computation tasks and lacks dynamic adaptability, whereas AI Agents typically handle varied tasks and complex contexts, which rigid architectures struggle to support.

In contrast, MetaArena’s zero-knowledge trusted execution solution is decentralized and does not rely on any centralized entity. Its security comes from a large-scale off-chain distributed computing network. This provides lightweight advantages and superior scalability and flexibility compared to TEE, enabling efficient adaptation to diverse AI Agent applications. MetaArena seamlessly supports LLMs like ChatGPT and trending models like DeepSeek. Notably, MetaArena’s solution is fully based on ZK cryptography, distinguishing it in the trusted AI Agent domain.

Overall, despite rapid AI technology iterations, full adoption of fully autonomous AI Agents still faces challenges in security, ethics, and practicality. Semi-autonomous AI Agents, balancing automation and human oversight, remain the mainstream. This means that before large-scale adoption, AI Agents require progress in trust and privacy, and MetaArena, with its fully ZK-based cryptographic solution, is accelerating this, laying a solid foundation for the next phase of AI Agent development. The new funding further establishes its position as a leading trusted AI engine infrastructure.

As one of the most important zero-knowledge trusted execution infrastructures in the AI era, MetaArena is “Making Agent Secure Again”!