As AI agents gradually gain execution rights on-chain, the trust paradigm is experiencing a fundamental shift — moving away from verifying signature authenticity toward verifying the alignment of inputs, decisions, and execution. This transformation is not merely theoretical; it's being forced into reality by genuine security disasters. When smart accounts and strategy wallets become mainstream interaction interfaces, a single valid cryptographic signature can only prove that an authorized path was invoked — it cannot guarantee that the data relied upon by the agent is reliable or that the decision aligns with the established strategy. The boundaries of verification are being redefined, extending from purely mathematical correctness to the truthful matching of logic and state.

The KelpDAO rsETH cross-chain bridge attack, which occurred on April 18, 2026, serves as a painful real-world validation of this paradigm shift. The incident led to the abnormal release of 116,500 rsETH, valued at approximately $292 million at the time. LayerZero's incident report revealed the complexity of the attack chain: the attacker first obtained developer session keys through social engineering, then contaminated the internal RPC relied upon by the LayerZero Labs DVN, and used denial-of-service attacks to suppress external RPCs, forcing the signing service to issue attestations for forged messages based on incorrect data. The more critical vulnerability lay in configuration errors — KelpDAO had downgraded its verification path from 2-of-2 to 1-of-1 DVN. This meant that once the sole designated DVN issued a flawed attestation, the system no longer required a second independent DVN for cross-verification, allowing a single point of failure to evolve directly into systemic collapse. CrowdStrike (CRWD.US) and Mandiant attributed the incident with high confidence to the North Korea-linked TraderTraitor (UNC4899).

Where the real risk now lives

Incidents of this nature demonstrate that the core contradiction in on-chain security has shifted from the robustness of cryptographic assumptions to engineering implementation details such as key control, data provenance, validator configuration, and operational permissions. The system must not only answer "is the signature valid," but also "who has the right to sign, based on what information are they signing, and did the state corresponding to the signature actually occur." Existing trust solutions all reveal significant limitations when confronting such complex scenarios.

In oracle and dispute resolution mechanisms, Polymarket's market outcomes are first proposed by participants and only enter UMA's token-holder voting arbitration if challenged during the challenge window. The core risk lies not in the absence of review, but in the reliability of that review — when rules are ambiguous or voting power is concentrated, the system effectively delegates the power to "define facts" to another governance structure. While cross-chain bridge multisigs and DVNs achieve this through different means, they all rely on application parties explicitly configuring their validator sets. The KelpDAO 1-of-1 DVN configuration serves as a textbook negative example — a single verification service's data sources and failure-handling mechanisms easily become a new single point of failure.

Although MPC custody achieves key sharding through threshold signatures, it does not automatically bring about organizational-level power decentralization. As disclosed by the Multichain team, after the founder was detained by Chinese police, the team immediately lost access to the related MPC node servers because those servers ran under the founder's personal cloud accounts. The centralization of cloud accounts, operational permissions, and emergency response leaves MPC's threshold design vulnerable at the organizational level. TEE (Trusted Execution Environment) can isolate code and sensitive data, but the trust anchor doesn't disappear — it merely transfers to hardware root of trust and microcode updates controlled by chip manufacturers, as well as enclave code and upgrade permissions controlled by project teams. TEE protects the computation process but cannot automatically decentralize governance authority.

Why just 10 ASX 200 shares?

These solutions have different failure modes, yet they converge on the same issue: the decentralization promises in white papers only constitute a real security boundary when implemented at the level of data sources, account permissions, and governance processes. DeepSafe, renamed from Bool Network in 2025, attempts to solve this trust dispersion problem through technological architecture innovation with its core product, CRVA. CRVA continues the technical approach proposed by Bool Network-affiliated researchers in 2022, with the related paper published in IEEE Transactions on Information Forensics and Security (IEEE TIFS, Document ID 9903072), proposing a cross-chain notarization platform based on an "evolving hidden committee."

In the specific mechanism, nodes participate in random selection through Ring-VRF, with selected nodes submitting proofs and temporary public keys — external observers can verify their eligibility but find it difficult to identify their long-term identities. The selected temporary committee then jointly signs through threshold MPC, ensuring no single node can independently produce results. Key management and other critical processes run inside TEEs (using Intel (INTC.US) SGX as an example) per the paper's design, aiming to reduce the possibility of host operators reading or tampering with key shares. Committees rotate by epoch, with each new committee receiving new shares through verifiable key handover while old shares become invalid. The project also intends to use TEE to hide committee working status, making it difficult for node operators to determine whether they themselves are participating in verification.

Data compiled by Woofun AI indicates this dynamic hiding mechanism aims to structurally weaken targeted attacks against fixed validators, though its actual effectiveness depends heavily on the maturity of production network code, remote attestation, and side-channel protection. While CRVA holds theoretical advantages in reducing exposure risks for fixed validators, its boundaries and risks remain clear. It cannot eliminate governance single points, software supply chain risks, or issues with the rules themselves. Node admission, protocol upgrades, TEE attestation, and software security all require ongoing auditing. Provided that old shares are reliably invalidated and new committees maintain sufficient independence, the rotation mechanism can shorten attack windows targeting fixed signing groups, but it cannot cover systemic risks such as software supply chain or upgrade permissions.

CRVA's technical lineage traces back to the Bool Network paper published in IEEE TIFS Volume 17 (2022), DOI 10.1109/TIFS.2022.3209546. The paper's protocol models, security proofs, and prototype evaluations underwent peer review, providing academic grounding for dynamic hidden committees, Ring-VRF, threshold key management, and TEE protection. However, a strict distinction must be drawn: peer review applied to the paper's models and implementations; the correspondence between DeepSafe's currently deployed CRVA and the paper's scheme still requires independent judgment based on current technical specifications, code audits, and network parameters.

CRVA addresses "who verifies and how results are securely jointly produced," but it does not automatically define "what results are correct." In AI agent scenarios, committees must still reach conclusions based on preset strategies, data sources, and judgment rules — if the rules themselves have defects or data sources are unreliable, even the most secure committee can collectively confirm incorrect conclusions. DeepSafe's ecosystem progress reflects market demand for trust restructuring under modular blockchain architectures. In October 2025, DeepSafe announced the completion of a $3 million seed funding round, with investors including Antalpha Global, ViaBTC Capital, and Gate. Data from the same period shows the network has cumulatively processed nearly 120 million verifications, with over 2.65 million active accounts and more than 70 ecosystem partnerships spanning wallet compatibility, technical integration, investment, and market collaboration. This funding round primarily corresponds to technical R&D and ecosystem expansion following the brand rename.

As blockchain infrastructure becomes increasingly modular — with consensus, execution, data availability, interoperability, and account systems handled by different components — trust issues do not disappear; rather, each layer's security boundaries become more clearly defined. Developers must not only choose technologies but also judge who provides security guarantees and where responsibility lies. As AI agents gain on-chain execution capabilities, confirming that data is trustworthy, decisions have not exceeded authority, and transactions match authorizations becomes the new challenge. DeepSafe's "Proof, Not Promises" vision aims to abstract verification capabilities from application-embedded modules into infrastructure that different protocols and AI agents can call upon. CRVA has combined anonymous selection, threshold collaboration, and TEEs into a technical pathway, yet whether it can evolve into universal verification infrastructure depends on continued accumulation of production network capabilities, independent audits, and real-world integrations. This represents a critical attempt in the evolution of Web3 trust architecture toward dynamic and anonymous mechanisms — following in the footsteps of traditional multisig and oracle systems.