MetaArena is a trusted execution infrastructure aimed at AI Agents and complex interactive systems, designed to accelerate the adoption of AI across various fields including gaming and finance. It ensures that intelligent behavior is verifiable, auditable, and scalable through zero-knowledge proofs, providing a genuinely trustworthy intelligent experience for blockchain games.

As one of the few infrastructure projects currently focused on trusted execution for blockchain gaming scenarios, MetaArena has recently garnered significant market attention and successfully completed a new round of strategic financing. This round saw participation from several well-known 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, A1E Omega, among others.

This round of financing not only represents the capital market's high recognition of the MetaArena narrative but also further validates MetaArena's potential value and strategic position in the upcoming wave of on-chain intelligent interaction upgrades.

MetaArena, how to make AI Agents more trustworthy?

MetaArena itself is a trusted execution infrastructure centered around zero-knowledge proof mechanisms, specifically designed to provide trusted computing services for AI games and intelligent interaction scenarios that require verifiable execution.

MetaArena includes an off-chain computing network composed of distributed computing nodes and an on-chain verification engine deployed in a multi-chain environment. When there are trusted execution tasks in the network, MetaArena will distribute AI behavior requests to off-chain computing nodes for execution and generate zero-knowledge proofs (ZKP) for them, followed by on-chain verification. Through this mechanism, it ensures that input data, inference behavior, and execution results are all authentic, trustworthy, and tamper-proof. MetaArena has been validated in the Web3 gaming field and supports AI Agent-driven on-chain games to operate efficiently, securely, and with auditing capabilities without relying on centralized servers.

In a recent upgrade, MetaArena launched a brand-new trusted execution stack, implementing verifications for prompt input consistency (Proof of Prompt) and inference behavior trustworthiness (Proof of Inference) through the two core capabilities of zkTrace and zkAction, to ensure the authenticity and confidentiality of prompts and inference paths during the execution of AI Agents in a provable manner.

It is worth mentioning that despite the many solutions currently on the market attempting to provide trusted operating environments for AI Agents, MetaArena is one of the few solutions that purely relies on zero-knowledge cryptography and does not require dedicated hardware to achieve trusted execution.

zkTrace: Proof of Prompt Input Trustworthiness

In traditional AI Agent models, a core problem has always been unsolved: how to ensure the trustworthiness of the prompt?
Including but not limited to:

• Was the prompt tampered with before or during execution?

• Does the model indeed infer based on the expected prompt?

• Is there a risk of sensitive content in the prompt being leaked?

MetaArena provides verifiable and trustworthy execution capabilities for prompts at the computational layer through the zkTrace module, ensuring that prompts maintain correctness, consistency, and privacy throughout their lifecycle; there is no need to expose their original content to the outside world, which is an important foundational component for building trustless AI Agents and decentralized application logic.

zkTrace is provided in a developer-friendly SDK format, underpinned by strong encryption mechanisms and ZK primitives, including Pedersen commitments / Poseidon / zkSNARKs (Plonk), and deeply collaborates with the System Prompt initialization process.
At system initialization, the prompt will be used as input to generate encrypted commitments via the off-chain computing network, constructing the corresponding ZKP.

These ZKPs can be referenced by any user or third-party verifier, comparing them with the on-chain proof of the prompt commitment to confirm the prompt's content is authentic and untampered. If the prompt used for execution is inconsistent with the audit commitment, verification will immediately fail, thereby ensuring behavior transparency and reliable execution without exposing plaintext.

In the usage process, AI Agent developers or AI prompt application developers can use zkTrace to create and define System Prompts to ensure that the model executes tasks strictly according to established strategies and constraints.

Once the System Prompt is initialized and loaded into the model, zkTrace will automatically generate commitments and proof files for it and submit them to the on-chain verification engine. This process fully records the complete trajectory of the prompt from input to use, ensuring that the proof is traceable and cannot be forged.

For end users using AI Agents, they can obtain the prompt commitments and proofs corresponding to the current execution model at any time and verify the authenticity of the prompt usage:

Does it still remain consistent with the developer's settings?
Has it been replaced or injected with malicious content during execution?

zkTrace ensures that trust in prompts no longer relies on centralized hosting or a single service provider's endorsement, but rather establishes a verifiable, auditable, and non-repudiable system input trust foundation through cryptographic proofs.

zkTrace Interaction Example

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

When AI Agent developers define and submit System Prompts through zkTrace, the prompts will be encrypted off-chain and generate commitments while completing agent initialization and binding to the corresponding verification circuits, ensuring that the prompt content has tamper-proof properties throughout the entire operating system. Meanwhile, the AI Agent will register the necessary verification keys to the MetaArena off-chain computing network for subsequent verification calls.

When the DApp initiates a message or interaction request, the AI Agent will read the request and delegate the execution task to off-chain computing nodes. During execution, the use of prompts and logic configurations will be verified through zero-knowledge proof mechanisms, and the behavior path will be recorded and generate verifiable proof files. The proof results will then be returned to the smart contract or DApp for contract-level confirmation that the behavior strictly originates from the promised prompt.

MetaArena's on-chain verification engine is responsible for matching and verifying zero-knowledge proofs and commitments to confirm the consistency of input content and execution behavior. If there are cases of prompt replacement or deviation in execution strategies, verification will immediately fail, effectively curbing potential anomalous behavior links. This mechanism ensures that the execution process of the AI Agent is fully aligned with the initially set parameters and has a transparent and auditable trust foundation.

By collaborating with smart contracts and other on-chain objects, MetaArena enables AI Agents to execute with publicly verifiable attributes, providing high security and structured trust for various Web3 use cases.

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

• Data Privacy: The content of prompts can be verified for authenticity without being made public, avoiding the leakage of sensitive information.

• Trustworthiness and Transparency: Zero-knowledge proofs ensure that the model's behavior has not been maliciously tampered with.

• Distributed Verification Capabilities: Any user or third party can verify execution consistency, avoiding reliance on centralized entities.

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

Overall, zkTrace is particularly suitable for critical task scenarios, such as intelligent decision-making tasks involving sensitive financial information, strict constraint policies, or high compliance requirements, building a highly secure and transparent operational foundation for the next generation of trustless AI Agents.

Trusted Framework for AI Agent Game Engine

MetaArena has taken the lead in the field of on-chain gaming, launching the AI Game Engine component, which allows the operations of agents in games to be constrained and audited by zero-knowledge proof mechanisms. Game agents can directly participate in on-chain battles through smart contracts, and their behavior is verified by zkTrace / zkAction to ensure fairness, authenticity, and traceability of the matches.

In this game engine system, developers can continue to use native game engines like Unity, Cocos Creator, and Unreal for development without changing existing creation methods to migrate games to a trusted on-chain operating environment. Developers can access MetaArena's decentralized state layer through SDK interfaces to manage key on-chain states, including player input, state changes, and round transitions, and complete verifications in real time through zero-knowledge proofs.

All generated content and task feedback can be handled by multiple AI Agents (such as content generation agents, game agents, testing agents, etc.), achieving automated verification and dynamic game experience optimization.

All data generated during the game process—including instruction inputs, state transitions, behavior logs, and content generation results—will be transmitted to MetaArena's off-chain computing network for processing and integrated into a verifiable proof structure through the ZK Game SDK. With the aid of ZK circuits (such as ZK Shuffle, action legality circuits), randomness, fairness, and rule consistency can be ensured. Meanwhile, the on-chain verification engine publicly confirms the authenticity and trustworthiness of each behavior through zero-knowledge verification mechanisms, ensuring that the game execution process is tamper-proof and completely transparent.

In terms of computation and storage, MetaArena combines resource optimization components to provide high-performance support for multiple agents (AIGC, QA testing agents, data insight agents, etc.), ensuring execution efficiency and response stability in high-throughput interactive scenarios.

Ultimately, this infrastructure not only provides developers with the efficient computing resources they need but also ensures that every step of game operations is verifiable, auditable, and accountable through a dual mechanism of decentralized verification and intelligent behavior auditing, thereby truly establishing a fair and trustworthy on-chain AI gaming ecosystem that effectively eliminates cheating, tampering, and covert execution.

Better Security

In the race to build trusted AI Agents, TEE solutions are widely adopted due to their hardware-isolated environments, which can achieve a certain degree of data privacy protection and execution verifiability. Although TEE is a mainstream privacy solution that has been validated and widely applied in multiple fields, it still has certain limitations in building trusted AI Agents.

In fact, TEE solutions typically rely on trusted environments and key management services provided by hardware manufacturers such as Intel SGX and ARM TrustZone. This centralized trust mechanism makes the system's security highly dependent on specific manufacturers, bringing concentrated risks; Intel SGX has previously been exposed multiple times for vulnerabilities that directly threaten its trusted foundation. Furthermore, although TEE provides an isolated execution environment, its data privacy protection capabilities still have shortcomings. For example, there may be eavesdropping risks during the transmission of data to the TEE environment, and external attackers may also access sensitive information through interaction interfaces. Additionally, TEE's design is primarily aimed at predefined computing tasks, lacking dynamic adjustment capabilities. AI Agents often need to deal with changing tasks and complex contextual scenarios, and this rigid architecture is also difficult to meet actual needs.

In contrast, MetaArena's zero-knowledge trusted execution solution possesses decentralized characteristics, not relying on any centralized entities, with its security stemming from a large-scale distributed computing network cluster off-chain. This not only gives it lightweight advantages but also significantly outperforms TEE in terms of scalability and dynamic flexibility, allowing it to adapt more efficiently to the diverse application scenarios of AI Agents. Whether it's ChatGPT or currently surging large language models like DeepSeek, MetaArena can achieve seamless compatibility. It is worth mentioning that the MetaArena solution is entirely based on ZK cryptography design, making it stand out in the field of trusted AI Agent solutions.

Overall, although AI technology is iterating and developing at an astonishing speed, under the constraints of security and ethical issues, as well as considerations of practicality, fully autonomous AI Agents still face many challenges to achieve widespread adoption. In contrast, semi-autonomous AI Agents, by balancing automation and human supervision, will remain the mainstream direction for future development. Similarly, this also means that AI Agents urgently need to make progress in trustworthiness and privacy before large-scale adoption, and MetaArena, with its entirely ZK-based cryptographic solutions, is accelerating this process and providing a solid foundation for the next stage of development for AI Agents, while the new financing is also laying the groundwork for its leading position as a trusted AI engine infrastructure.

As one of the most important zero-knowledge trusted execution infrastructures in the AI era, MetaArena is here to 'Make Agent Secure Again'!