Bridging UAE-based RWA structuring & compliance with on-chain governance infrastructure for DAOs.

DeXe Protocol is announcing a strategic partnership with RWAlabs.ae, a UAE-based full-scope provider supporting real-world asset tokenization — from regulatory structuring to execution and go-to-market support.

This partnership connects two complementary layers of the RWA stack:

• RWAlabs.ae will support asset tokenization programs with structuring, legal, and compliance rails in the United Arab Emirates.
• DeXe Protocol provides decentralized governance infrastructure for creating and managing DAOs — enabling transparent rules, on-chain decision-making, and treasury operations around tokenized assets.

RWA Tokenization: A Rapidly Scaling Market

RWA tokenization is increasingly shaped by structure, governance, and regulatory recognition — the practical foundations required for real assets to operate credibly on-chain at scale.

The market trajectory supports this shift: RWA.xyz reports $24.72B in distributed asset value and $344.09B in represented asset value across tokenized RWAs, with 844,216 holders tracked.

McKinsey & Company estimates tokenized market cap across asset classes could reach ~$2T by 2030 (excluding crypto and stablecoins).

A Full-Cycle Path for Tokenized RWAs

RWA Labs supports tokenization programs across a wide range of real-world assets — providing end-to-end structuring, legal, and compliance rails for execution in the UAE. In this strategic partnership, DeXe Protocol becomes the governance infrastructure layer around these programs, where tokenized assets can be managed through DAOs with transparent rules, on-chain decision-making, and treasury operations.

With RWA Labs as the structuring and compliance provider, DAOs built on DeXe gain a clearer path to a full-cycle setup — from tokenization through ongoing governance and asset management.

“Tokenization programs succeed or fail on fundamentals: structure, governance, and regulatory recognition. This partnership brings those rails together — linking robust RWA structuring with on-chain governance infrastructure that can coordinate ownership and decision-making transparently.” — Juliet Su, Founding Member, RWAlabs.ae

About DeXe Protocol

DeXe Protocol is a decentralized governance infrastructure for creating and managing DAOs. It serves as a universal governance layer to power on-chain collaboration for humans and AI agents, providing advanced solutions for building functional, transparent DAOs and implementing efficient decentralized governance mechanisms.

About RWA Labs (RWALabs.ae)

RWAlabs.ae is a UAE-based full-scope provider supporting asset tokenization programs, including company formation and legal structuring, alongside execution support.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

Blockchain promised transparency — a ledger where every transaction, every vote, every decision is permanently visible. This radical openness seemed like the path to trustlessness: no hidden deals, no backroom negotiations, no opacity.

But transparency has a dark side. When everyone can see how you vote, voters become targets. Early voters influence late voters, not through argument but through social pressure and herding. Token holders large enough to move markets can pressure token holders small enough to move. Bribers can confirm that a voter actually cast the bribe-demanded vote. Democracy becomes not a deliberative process but a theatre of visible coercion.

The paradox deepens: the more transparent the system, the less free the voting becomes.

A new cryptographic primitive offers an elegant escape from this trap: zero-knowledge proofs. They allow a voter to prove that their vote is valid — that they were eligible, that they voted once and only once, that their vote was counted — without revealing what they voted for or who they are. Transparency without surveillance. Verification without exposure.

The Bribery Problem: Transparency’s Hidden Cost

Consider a token holder with 1,000 governance tokens in a major DAO. A wealthy actor wants to influence a governance proposal. They approach the token holder with an offer: vote “yes” on this proposal, and I’ll pay you 100 ETH. But how can the briber trust the token holder will actually vote as promised?

In traditional voting systems, this is difficult. The vote is secret; the briber can’t verify. But in transparent DAOs, the vote is on-chain, permanent, and linkable to the voter’s address. The briber can simply check the blockchain and confirm the vote was cast. The threat of bribery is suddenly credible.

This isn’t theoretical. MakerDAO governance forums have documented cases where large holders were directly solicited to vote in exchange for payments. The fact that voting is publicly visible on-chain makes such arrangements enforceable. The briber doesn’t need to trust the voter; the blockchain enforces the contract.

Similar dynamics emerge with voter pressure. Large token holders can publicly signal their votes and create herding behavior. Not through deception, but simply by moving first. A whale votes “yes” on a proposal, and smaller holders follow because they assume the whale has done research. The voting choice becomes less about the merits of the proposal and more about following visible leaders.

Zero-Knowledge Proofs: Proof Without Exposure

Zero-knowledge proofs are a cryptographic construction that allows one party (the prover) to convince another party (the verifier) that a statement is true without revealing any information beyond that the statement is true.

For voting, this means:

• A voter can prove they held tokens on a snapshot date (eligible to vote) without revealing their address or token balance
• A voter can prove they cast exactly one vote without revealing which option they voted for
• A voter can prove their vote was counted without revealing the content of their vote
• The final vote tally can be cryptographically proven correct without revealing how any individual voted

The magic of zero-knowledge is that you substitute evidence for exposure. The blockchain doesn’t need to see your vote; it needs to see a mathematical proof that your vote was valid and counted correctly. These proofs are typically far smaller than the original data, easier to verify, and reveal nothing about the voter’s choice or identity.

Real-World Implementation: MakerDAO’s Anti-Collusion System

MakerDAO governance became a laboratory for ZK voting precisely because of observed bribery and collusion. In 2023, researchers noted that large token holders could extract premium prices for their votes, and smaller holders faced pressure to follow visible voting patterns.

MakerDAO introduced zero-knowledge voting for sensitive parameter decisions. The mechanism works as follows:

1. A voter generates a cryptographic proof that they hold the required tokens on the snapshot block
2. The voter encodes their actual vote choice (encrypted)
3. The voter submits only the proof and encrypted vote to the blockchain, not their address
4. The blockchain verifies the proof (confirming voting eligibility without revealing identity) and accepts the encrypted vote
5. After the voting period closes, votes are decrypted and tallied
6. The final result is verified via ZK proof (proving the tally is correct without replaying individual votes)

The result was striking. When MakerDAO switched sensitive votes to ZK voting, voting patterns became significantly more independent. Voters were no longer clustered around visible whale decisions. The proposal outcomes showed greater diversity in reasoning and lower correlation with early voting signals. The governance process became less a theatre of visible preference and more an actual deliberation of ideas.

More concretely: bribery attempts dropped. Potential bribers have no way to verify that a voter actually cast the bribed vote, because votes are anonymous. The enforcement mechanism — checking the blockchain — is removed. Without verifiability, bribery loses its appeal.

Beyond DAO Voting: The Broader Infrastructure

The innovation extends beyond voting. Zero-knowledge proofs create a general-purpose layer for credible privacy. You can prove compliance with rules (your vote is in the set of allowed options) without revealing the details. You can prove membership in a group (you hold governance tokens) without revealing your balance or transaction history.

This opens doors to sophisticated governance mechanisms:

• Conviction voting with privacy: Vote your preference, prove your token-lock period, and remain anonymous
• Sybil-resistant mechanisms: Prove you are one person without revealing your identity
• Composable governance: Prove you’re part of a voting coalition without exposing the coalition’s members
• Sensitive risk parameters: DAOs can gather input on market manipulation strategies without broadcasting the vulnerabilities to attackers

Each of these benefits from the same core property: you can prove a statement without making it public.

The Trade-Offs: Verification as a New Burden

ZK voting is not a panacea. It introduces new complexities:

1. Computational cost: Generating and verifying zero-knowledge proofs requires cryptographic work. These proofs are faster than in the past, but not free. Voting becomes slightly more expensive.
2. Auditability concerns: Traditional voting leaves a voter with a receipt — they can confirm their vote was counted. With ZK voting, the voter knows they submitted a proof, but can’t independently verify the entire flow. This trades voter-side verifiability for voter-side privacy.
3. Trust in the setup: Most ZK systems require a “trusted setup” — a one-time ceremony where initial parameters are generated. If that setup is compromised, the entire system could be broken. Newer proof systems (like zk-STARKs) avoid this, but are less efficient.
4. Complexity as a barrier: ZK voting is less intuitive than traditional voting. A voter doesn’t “see” their vote on-chain; they see a proof. This reduces the psychological sense of verification that blockchain promised.

Despite these trade-offs, the evidence from MakerDAO and other DAOs is persuasive: privacy-preserving voting enables fairer governance by removing the credibility of coercion.

Privacy as Governance Infrastructure

The deeper insight is that privacy and transparency are not opposites in governance; they are complementary layers. You want:

• Transparency on the rules, the eligible voters, and the tally
• Privacy on individual choices and identities
• Verifiability that the process was correct

Zero-knowledge proofs bridge all three. They make governance simultaneously more open (anyone can verify the tally) and more private (no one can link votes to identities or coerce voters).

Toward Verifiable Privacy

Toward verifiable privacy, zero-knowledge voting shows that transparency and privacy need not be in conflict: governance can be both anonymous and auditable simultaneously. In practice, this does not always require heavy cryptography; it is often enough that individual votes remain hidden until the proposal closes, so no one can reliably monitor or enforce how a specific address voted during the process.

DeXe’s governance architecture follows this pattern exactly. Votes are fully recorded and auditable, but individual choices only become visible after the voting period ends, which makes bribery and direct pressure structurally hard to sustain while keeping outcomes and procedures transparent. It is a simpler form of verifiable privacy that addresses many of the same vulnerabilities highlighted in zero-knowledge research, without turning voting itself into a cryptographic ceremony.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the concepts proposed in “Governance-as-a-Service: A Multi-Agent Framework for AI System Compliance and Policy Enforcement” by Gaurav, Jukka Heikkonen, Jatin Chaudhary.

As AI agents evolve from tools into independent decision-makers, traditional governance mechanisms are showing their limits. Embedded filters and fine-tuning offer partial control but cannot keep pace with systems that continuously learn, adapt, and interact. The authors of Governance-as-a-Service (GaaS) propose a decisive shift: treating governance as an external, programmable layer that governs them in real-time.

Externalizing Control

Instead of modifying every agent, GaaS introduces an enforcement service that mediates between agents and their environment. Each action passes through this layer, where declarative policies decide whether to allow, warn, or block execution. The service updates a longitudinal Trust Factor for every agent, measuring its reliability over time. Governance becomes modular — applied from the outside rather than built into the agent’s code.

Three Modes of Oversight

The GaaS model defines three complementary modes of regulation. Coercive policies block actions that violate defined rules. Normative policies issue warnings for borderline behavior, creating feedback without halting the workflow. Adaptive policies escalate responses for repeated or severe violations. Together, they create a living feedback loop that shapes agent behavior through consequence rather than compliance alone.

Trust as a Dynamic Variable

A key innovation of GaaS is the introduction of the Trust Factor (TFa). Instead of binary approval, agents earn or lose trust based on their behavior, context, and the severity of the rule. TFa is updated automatically at runtime and can serve as a prerequisite for future privileges. The more consistent an agent’s compliance, the more autonomy it gains. Trust becomes quantifiable infrastructure — a governance primitive rather than a moral abstraction.

From Enforcement to Architecture

In simulations across content generation and financial automation tasks, GaaS successfully filtered harmful outputs and risky operations while maintaining high system throughput. Policies were implemented as machine-readable JSON files, and every decision was logged with rule IDs, timestamps, and contextual metadata. The result is a transparent, auditable layer of control that can evolve independently of any specific model or platform.

A logical next step is the creation of “Agent Passports” — verifiable credentials linking an agent’s Trust Factor, policy history, and access permissions. These passports could function like on-chain identities within the ecosystem, granting or restricting privileges automatically based on cumulative reputation. Delegation and validator layers would provide human oversight when context-sensitive interpretation is needed, maintaining a human-in-the-loop governance loop.

Governance as Shared Infrastructure

The lesson of GaaS is that governance can scale only when it becomes infrastructure. Compliance, reputation, and policy management should exist as services — transparent, composable, and collectively controlled. The enforcement layer need not be owned by any single actor; it can be maintained by DAOs that coordinate upgrades, manage policies, and ensure fair participation across agent communities.

DeXe’s architecture provides the foundation for such an ecosystem. By merging off-chain enforcement logic with on-chain decision-making, governance can evolve from a static rulebook into a dynamic, verifiable, and participatory economy. The result is not simply safer AI — it is accountable autonomy.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the ideas proposed in “From Cloud-Native to Trust-Native: A Protocol for Verifiable Multi-Agent Systems” by Muyang Liu.

As AI agents become autonomous and pervasive, the question is no longer how powerful they can get, but how verifiable their actions are. The transition from cloud-native to trust-native marks an evolution from systems optimized for availability to systems optimized for accountability and trust.

The authors of From Cloud-Native to Trust-Native argue that trust must become a design primitive. Their proposed framework, TrustTrack, provides a minimal yet complete protocol stack for verifiable multi-agent systems. It replaces blind faith with cryptographic evidence, defining how autonomous systems identify themselves, declare intent, and prove compliance — without undermining privacy or performance.

The New Bottleneck: Verifiability

Cloud-native infrastructure gave us elasticity, speed, and scale. However, when autonomous agents make financial, legal, or governance decisions, these virtues are insufficient. In cross-organizational and decentralized contexts, trust becomes the limiting factor.

Traditional observability tools can tell us what a system did, but not whether it should have done it. API logs, telemetry, and post-hoc audits fail when actors operate across boundaries with conflicting incentives. What’s missing is structural verifiability: an architecture that makes it cryptographically impossible to act outside agreed rules.

The TrustTrack Stack

The TrustTrack protocol formalizes this need into three composable layers: identity, policy, and behavior.

1. Agent Identity

Each agent possesses a decentralized identifier (DID) and cryptographic keys that bind its identity to metadata, such as model version, organizational ownership, or training provenance. This prevents anonymous execution and allows role-based accountability across systems.

2. Policy Commitments

Before acting, agents publish machine-readable policy commitments — formal declarations of allowed behaviors, constraints, and delegated authorities. These commitments are immutable, timestamped, and auditable, creating a shared ground truth for what each agent is supposed to do.

3. Behavioral Logging

Every action — input, decision, or output — is signed and hashed into a tamper-evident trace. Logs are batched into Merkle roots and periodically anchored to a blockchain or verifiable ledger. Sensitive details remain off-chain, while cryptographic proofs preserve integrity.

Together, these layers transform “trust me” into “prove it.” They create a programmable substrate for accountability, where agents carry not only intelligence but also evidence.

Why Protocols Matter More Than Policies

Legal or institutional regulation struggles to govern autonomous agents. It is reactive, jurisdiction-bound, and cannot scale to transnational digital systems. By contrast, protocol-level governance — the enforcement of rules through cryptography, incentives, and consensus — works natively in multi-agent environments.

This idea echoes a broader shift toward regulation by architecture: instead of building compliance after the fact, we embed it into the operational fabric. In trust-native systems, governance is not an overlay but a built-in property.

Selective Privacy and Incentivizing Provenance

A frequent objection to full verifiability is the loss of privacy. But TrustTrack and DeXe share a more nuanced stance: privacy and accountability can reinforce each other.

Zero-knowledge proofs and selective disclosure allow systems to demonstrate compliance without revealing the underlying data. This creates a privacy-preserving audit trail where sensitive information stays encrypted, but integrity remains provable. It’s a practical middle ground: agents stay accountable without being exposed.

Every autonomous decision has upstream contributors: data curators, policy designers, and prompt engineers. Trust-native architecture allows their efforts to be traced and rewarded. By integrating provenance proofs into DAO treasuries, we can compensate contributors whose artifacts demonstrably improve agent reliability. This transforms governance from cost to collaboration, making stewardship an economic role, not just an ethical one.

Toward a Trust-Native Future

The next generation of systems will not be judged by how fast they act, but by how verifiably they align with shared norms. The convergence of cryptography, decentralized governance, and verifiable AI offers a blueprint for trustworthy autonomy.

Autonomy without accountability breeds opacity. Accountability without autonomy breeds stagnation. Trust-native design bridges the two: allowing intelligence to act freely, but never without proof.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the ideas proposed in “DAO-AI: Evaluating Collective Decision-Making through Agentic AI” by Agostino Capponi, Alfio Gliozzo, Chunghyun Han, and Junkyu Lee (Columbia University & IBM Research, 2025).

As decentralized governance expands, one paradox becomes increasingly visible: the promise of collective intelligence collides with the limits of human attention. In most DAOs, fewer than 10% of token holders participate in votes. The rest remain silent, not out of indifference but exhaustion. Proposals grow complex, forums overflow with information, and deliberation struggles to keep pace.

The researchers behind DAO-AI built an agentic system capable of reading governance proposals, analyzing community discussions, and forming its own interpretable position. The result is not an oracle but an analytical mirror — an AI that reflects the logic of the collective and tests its coherence.

Reading, Reasoning, Voting

DAO-AI is built on the IBM Agentics framework, where information is modeled through typed data structures called ATypes. These entities represent text, metrics, or contextual signals — from forum posts to on-chain activity — connected through logical transductions that capture meaning.

Modular Composable Programs (MCPs) assemble these signals into reasoning pipelines. They fetch proposal metadata, aggregate discussion dynamics, analyze historical patterns, and assess market responses. The system then issues a decision — for, against, or abstain — accompanied by an explanation and a confidence score.

Across 3 383 proposals from DAOs such as Aave, Lido, and Uniswap, the AI’s votes aligned with collective outcomes in roughly 92–93 percent of cases. Even in contested situations, where human opinion was sharply divided, DAO-AI displayed consistency and interpretability. Its recommendations also correlated with favorable post-vote token and protocol performance, suggesting that agentic reasoning can echo not only sentiment but underlying rationality.

The Limits of Machine Alignment

The study’s authors are careful not to overstate their findings. DAO-AI does not prove that artificial agents make better decisions; it only demonstrates that they can approximate and explain collective reasoning at scale. The model was trained on a single configuration and a finite data horizon; causality remains untested.

Future research aims to extend coverage to additional protocols, develop contested-case benchmarks, and integrate human-in-the-loop audits to evaluate when and why the agent’s logic diverges from community outcomes. Over time, this approach could evolve toward causal assessment — measuring whether agentic participation genuinely improves governance quality.

From Automation to Translation

What makes DAO-AI remarkable it transforms dispersed deliberation into structured, interpretable reasoning. Instead of replacing voters, agentic systems could act as transparent advisors, contextualizing each decision within prior outcomes and the ethical trade-offs that have been made.

Over time, such systems may evolve into delegate passports — dynamic profiles where both humans and agents earn credibility through their alignment with collective outcomes. Reputation becomes a measure of coherence, rather than conformity. Advisors guide without obscuring accountability; delegates represent not authority but continuity of reasoning.

By bridging discussion data, on-chain evidence, and economic signals, agentic AI shifts governance from ritual to reflection. It helps communities reason in public — where every recommendation is auditable, traceable, and open to dispute.

A New Kind of Collective Reason

The broader implication of DAO-AI is clear: the next phase of decentralized governance will depend on collaboration between human judgment and synthetic reasoning. The task is not to surrender decisions to machines but to design frameworks where both kinds of intelligence reinforce transparency and shared purpose.

When collective decision-making becomes interpretable, governance shifts from a process of counting votes to a practice of understanding why votes were cast. Agentic systems of the future — one where deliberation scales without losing its conscience.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the concepts proposed in “The Trust Fabric: Decentralized Interoperability and Economic Coordination for the Agentic Web” by Sree Bhargavi Balija, Rekha Singal, Ramesh Raskar, Erfan Darzi, Raghu Bala, Thomas Hardjono, and Ken Huang.

As AI agents transition from demos to production, the dream of an Agentic Web, a global substrate where autonomous systems discover, collaborate, and transact, faces a stubborn obstacle: fragmentation. Today, protocols such as the Model Context Protocol (MCP), Agent-to-Agent (A2A), and Agent Collaboration Protocols (ACP) enable agents to communicate with each other. But communication alone does not create trust, accountability, or sustainable economic coordination. Without shared standards for identity, behavior, and incentives, agent ecosystems risk becoming silos — interoperable in name, but brittle in practice.

Networked AI Agents in Decentralized Architecture

The Networked AI Agents in Decentralized Architecture (NANDA) proposes a comprehensive five-layer model to stitch these pieces into a coherent fabric.

Layer 1: Discovery

The Where. Agents are registered through decentralized identifiers (DIDs) and verifiable credentials in federated registries that adhere to trust policies. A Learn-to-Rank model removes duplicates and ensures that relevant results are returned.

Layer 2: Composition

The How. Coordination is enabled by Agent-to-Agent (A2A) and Agent Communication Policy (ACP) protocols. IO Mappers translate between ontologies, enabling heterogeneous agents to interact and communicate effectively.

Layer 3: Deployment

The Runtime. Agents run in sandboxed containers, such as WebAssembly, ensuring secure execution across cloud, edge, or local nodes without compromising host systems.

Layer 4: Evaluation

The Trust Engine. Trust scores are calculated based on policy compliance, behavioral telemetry, and cryptographic attestations. Scores adapt to context, rewarding reliable agents and constraining risky ones.

Layer 5: Incentivization

The Why. Micropayment protocols (X42, H42) enable task-based, pay-per-use rewards. Contributions are compensated directly within execution flows, sustaining cooperation without pre-existing trust.

These layers create an operational fabric where agents can be discovered, coordinated, executed, evaluated, and incentivized in a unified system.

Trust as the Core Layer

What sets NANDA apart is the centrality of trust. Instead of assuming agents will behave responsibly, the framework continuously measures, verifies, and updates their trustworthiness. Trust scores are built from policy compliance, behavioral telemetry, and cryptographic attestations, adapting to different contexts and tasks. This transforms trust into a programmable asset: agents with higher scores gain more opportunities, while poor performance results in restrictions or exclusion. In effect, trust becomes the currency of coordination: auditable, dynamic, and tied directly to economic outcomes.

From Blueprint to Practice

The framework points to practical domains where its five layers already matter. In emerging agent marketplaces, credentialed agents are discoverable, ranked, and paid directly for services like data access or inference tasks. In healthcare, compliance cannot rely on promises: verifiable credentials and cryptographic attestations ensure that only certified agents can handle sensitive patient data. In enterprise settings, interoperability becomes critical: IO Mappers and federated registries allow agents from different vendors to collaborate without exposing proprietary systems.

By embedding trust and incentives at the protocol level, NANDA shows how these examples can move beyond isolated pilots into scalable, production-ready ecosystems. Instead of ad-hoc integrations, agent economies gain a common backbone where discovery, evaluation, and reward are baked into the architecture itself.

DeXe’s Alignment

What stands out in the NANDA framework is how governance is not treated as an add-on but as a structural necessity. The Trust Engine turns reliability into a measurable resource that regulates access, incentives, and collaboration, making the ecosystem accountable and adaptable by design. This vision aligns with the idea that governance is the layer that binds discovery, execution, and incentives into real-world systems. Trust as a programmable foundation enables agent networks to remain open, scalable, and ethically grounded, which is exactly the direction being pursued in decentralized governance today.

Toward a Trusted Agentic Web

The Agentic Web will not thrive solely on protocols. It requires a fabric of trust and incentives to bind autonomous actors into coherent, transparent, and sustainable networks. The Networked AI Agents in Decentralized Architecture provides the blueprint for such a system, showing how discovery, execution, evaluation, and economics can operate as one fabric.

By balancing research, commercial application, and community governance, autonomous agents can evolve from isolated tools into interconnected participants in a trustworthy digital economy. The task ahead is not only to design these systems but to ensure they remain open, accountable, and aligned with human values.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the concepts proposed in “Virtual Agent Economies” by Nenad Tomašev, Matija Franklin, Joel Z. Leibo, Julian Jacobs, William A. Cunningham, Iason Gabriel, and Simon Osindero.

The rise of autonomous AI agents marks a turning point: from tools that execute commands to economic actors that transact, negotiate, and coordinate. The paper Virtual Agent Economies frames this shift and asks a pressing question: how do we design these markets so they remain steerable before they lock in as critical infrastructure?

When autonomous agents interact directly, they form what the authors call sandbox economies, bounded environments where agents exchange services, data, and compute. To classify these sandboxes, the paper introduces a simple 2×2 framework: whether they are emergent or intentional, and whether they are impermeable or permeable to the human economy.

Emergent, highly permeable sandboxes may accelerate innovation, but they also create serious risks by tightly coupling machine dynamics with real-world markets. Intentional and less permeable sandboxes, by contrast, give society more control and a safe place to test mechanisms. The critical design variable is permeability, and it is a collective property, not a lever a single actor can pull.

Building a Trust Stack

Trust in these systems cannot be assumed. It must be verifiable and portable. The authors highlight a trust stack made up of several key components. Decentralized Identifiers provide a base layer of identity. Verifiable Credentials carry records of an agent’s history, capabilities, or access rights. Zero-Knowledge Proofs allow agents to prove compliance with rules without revealing raw data. And Proof-of-Personhood adds Sybil resistance whenever fairness requires mapping agents back to unique human users.

Together, these tools transform trust into a measurable asset, making cooperation economically rational while keeping interactions private and secure.

Incentives and High-Frequency Negotiation

Once trust is established, incentives ensure participation. Why should an agent contribute resources instead of free-riding? The paper points to auction mechanisms as a way to allocate scarce resources like compute or data access fairly, coupled with reputation systems that reward proven reliability.

The authors also highlight a phenomenon they call High-Frequency Negotiation (HFN). In the near future, personal AIs may negotiate continuously on behalf of users, striking countless micro-deals at speeds humans cannot match. Like high-frequency trading, this could make systems efficient, but it also risks amplifying advantage for well-equipped actors and creating systemic instability. Oversight and pacing mechanisms will be essential.

Mission Economies and Community Currencies

Beyond generic markets, the paper introduces the idea of mission economies, where sandboxes are dedicated to solving global challenges such as scientific research, climate adaptation, or health. These markets are powered by community currencies that localize incentives, keeping cooperation tied to collective goals instead of speculation.

By adjusting transferability, velocity, and scope, designers can tune these currencies to channel resources where they are needed most. It is a way to embed alignment directly into the economy, not just into governance rules.

Containment and Oversight

With great speed comes great fragility. The authors argue that safe deployment will require regulatory sandboxes and strong containment mechanisms. Oversight agents should have the power to quarantine misbehaving actors. Transparent audit trails and layered escalation ensure that interventions are both automatic and accountable.

Crucially, these ideas are not left abstract. The paper walks through scenarios where sandbox economies could prove transformative: marketplaces for scientific collaboration, coordination across robotic fleets, and personal assistant networks that handle service negotiation at scale. In each case, the opportunities are immense — but only if governance is built in from the start.

DeXe and Virtual Agent Economies

The framework laid out in Virtual Agent Economies resonates strongly with DeXe’s design philosophy. Our governance stack is already engineered around programmable rules, modular incentives, and reputation-aware participation, the very building blocks the paper identifies as essential.

Through metagovernance architecture, DeXe can host intentional sandboxes where missions are clearly defined, permeability is consciously managed, and local currencies keep cooperation aligned. Non-linear, meritocratic voting prevents capture by wealthy actors, echoing the call for fairer participation. And our transparent governance processes ensure that experimentation remains auditable and accountable.

DeXe does not pretend to solve every challenge, but it offers a live environment where these ideas can move from research to practice. In this sense, the protocol acts as a proving ground for the design of steerable agent markets.

Toward Steerable Autonomy

The message of Virtual Agent Economies is clear: the time to design is now. Agent markets will not wait for regulation to catch up; they are already forming. If we want them to serve human goals, we must embed verifiable trust, aligned incentives, and robust oversight before they ossify into the default infrastructure of digital life.

DeXe’s contribution is to make this design space real, providing programmable, modular environments where communities can test, adapt, and refine the governance of agent economies. Done right, these markets can be both autonomous and accountable, delivering innovation at machine speed without abandoning human steerability.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

The AI market is rapidly transforming Web3: from content generation to financial decisions and agent-based products. Power scales up, and today the key question is trust. Who determines how these systems behave, and under what rules can they be corrected? Centralized providers deliver speed, but bring opacity, single points of failure, and arbitrary policy switches.

In an environment that values open rules and verifiability, that’s not enough.

That’s the reason ChainGPT and DeXe are joining forces. ChainGPT contributes a mature AI stack for Web3: generation, analytics, APIs, and smart contract auditing. DeXe is a decentralized governance protocol that turns AI ownership into a transparent, collective, accountable primitive, unlocking secure human-agent collaboration in programmable economies.

This provides a solid foundation for a governable deAI architecture, where intelligence is backed by transparent governance.

Why now

We’re in an AI investment super-cycle: major players are pouring cash flow back into GPUs, data centers, and power, stitching the stack from silicon to serving. As agents move from demos to production and start touching capital and distribution, they create a governance gap unless they ship value with verifiable controls.

What the market needs now is transparent AI operations: clear lineage of how models are trained and updated, provable runs and accountable behavior, policy enforced as code at execution time, and tamper-evident logs for audit.

Web3 brings open state, programmable incentives, and neutral enforcement, turning policy into contracts and alignment into markets, so transparent training and accountable inference become table stakes.

What we’re connecting

Decentralized AI has such significant bottlenecks: provenance, quality verification, incentives and accountability, and agent coordination. To address them, the build splits cleanly: ChainGPT is shipping AIVM, a purpose-built Layer-1 for decentralized AI (testnet targeted for late 2025), designed to run agents and verifiable inference at the protocol level; DeXe provides robust governance, incentivization, reputation, and automation: treasury management, voting, roles, delegation, on-chain logs, incentives, and meta-governance structures. Together, this composes a three-layer system.

At the bottom is the intelligence layer: models and agents generating and analyzing artifacts, operating through APIs and developer tools. In this layer, we explore verifiable content flows and agent reputation, so artifacts and actions carry provenance and performance history by default.

At the top is the governance layer: roles, reputation, voting, escalation rules, on-chain logs, and treasury mechanics. Human-in-the-loop and ethics-as-code crystallize as roles, rule updates, and transparent model cards, keeping alignment programmable rather than aspirational.

In between is the execution and audit layer: smart contracts that translate decisions into actions (permissions, slashing, policy updates), and logs that capture the lifecycle of artifacts and agent behavior. Here, reliability incentives and dispute mechanics settle outcomes on-chain, with policies enforced as code and tamper-evident logs captured end-to-end.

Areas of joint work

After preliminary joint research, we defined where combining efforts delivers the most significant impact and where ecosystem feedback is crucial. In these five tracks, AIVM could provide significant value:

Decentralized validation of AI content

When it comes to news, research, or analytics, quality is the validation procedure. Imagine flows where the community acts as a distributed editor: selective pre-moderation, validator curation, disputes, and corrections with an on-chain trace. Concretely, on-chain flows log claims, sources, and fixes, rewarding accuracy over noise and keeping provenance queryable.

On-chain reputation for AI agents

Every agent gains a history: what it has done, how accurate it was, and how quickly it fixed errors. Reputation updates algorithmically and is reinforced by economics, staking roles, and slashing for harm. Agents can maintain verifiable scorecards: accuracy, responsiveness, time-to-fix, anchored on-chain, and portable across apps.

Incentivization mechanisms for AI reliability

Research how decentralized incentive structures can align participants and AI agents toward transparent, accountable, and self-correcting behavior. Economic designs reward correct outputs and penalize low-quality disputes, suppressing noise while elevating high-confidence results.

Human-in-the-loop through governance

User feedback routes into working groups, requires reasoned decisions, and, where possible, auto-applies via rules. Governance primitives translate community signals into enforceable on-chain rules with clear upgrade paths and circuit breakers.

An ethical framework for Web3 contexts

Working on a conceptual framework for “context-aware AI ethics” based on the decentralized ecosystem’s programmable economy and governance principles, where ethics becomes the interface between values and execution. Policy-as-code and transparent model cards bind domain constraints to behavior so decisions are auditable and context-aware.

As ChainGPT’s AIVM moves toward testnet and beyond, we’ll continue to explore these tracks and the potential to contribute meaningfully across them.

ChainGPT’s Research & Development Focus Areas

ChainGPT’s AIVM (AI Virtual Machine) is a purpose-built layer-1 for decentralized AI, scheduled for testnet release in late 2025. Alongside the core build, ChainGPT is contributing research and development across several areas where AIVM could provide significant value:

• Decentralized validation of content — On-chain flows could log claims, sources, and corrections, rewarding accuracy over noise.
• On-chain reputation for AI agents — Agents might build verifiable scorecards of accuracy, responsiveness, and fixes.
• Incentivization for reliability — Economic structures could reward correct outputs and penalize low-quality disputes.
• Human-in-the-loop governance — Community feedback might translate into enforceable, on-chain rules.
• Ethical framework for Web3 — Policy-as-code and transparent model cards could embed ecosystem values into execution.

As AIVM moves toward testnet and beyond, ChainGPT will continue to explore these areas and the potential for the protocol to contribute meaningfully across them.

Agent economy: the next layer of Web3

Agents are becoming full economic participants: they trade, make decisions, and carry out missions. We see this as a governable and trustworthy framework: intelligence meets governance, and from this emerges a rule layer where agents act transparently and predictably for people and protocols.

In practice, it’s one principle: intelligence creates value, governance makes it verifiable and scalable. Everything else is an implementation detail that can evolve without breaking the premise. Together, we lay the foundation for the programmable AI economy in Web3, where value is created fast, and the creation rules are as clear as a smart contract’s math.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

An exploration of the concepts proposed in Agentic Web: Weaving the Next Web with AI Agents by Yingxuan Yang, Mulei Ma, Yuxuan Huang, Huacan Chai, Chenyu Gong, Haoran Geng, Yuanjian Zhou, Ying Wen, Meng Fang, Muhao Chen, Shangding Gu, Ming Jin, Costas Spanos, Yang Yang, Pieter Abbeel, Dawn Song, Weinan Zhang, and Jun Wang (2025).

If interoperability ensures that agents can talk to each other, composability ensures that they can build together. The defining strength of the Agentic Web is not individual agents executing isolated tasks, but clusters of agents combining into higher-order services. Just as smart contracts in DeFi unlocked financial Lego, agentic composability unlocks the possibility of service Lego, where autonomous actors coordinate, delegate, and co-create across domains.

Composability as Principle

In this vision, agents are not endpoints but modules. A research assistant can combine with a trading bot, a logistics coordinator, and a compliance auditor to form an ad hoc team. Each retains autonomy, but together they deliver outcomes no single agent could achieve.

Composability shifts the paradigm from isolated tools to dynamic ecosystems of collaboration. Agents cease to be “apps” in the old sense and instead become interoperable services, composable into workflows as diverse as financial risk management, scientific research, or global supply chain optimization.

From Local Rules to Global Coherence

Yet composability introduces a scaling problem. Local clusters of agents may operate under their own standards, governance models, and reputation systems. One cluster may prioritize speed, another compliance, and a third cost-efficiency. These local logics work in isolation but can collide when clusters interact, producing contradictions or vulnerabilities at scale.

Without mechanisms for alignment, the Agentic Web risks replicating the very fragmentation it seeks to overcome: silos of agent clusters unable to integrate into a coherent whole. The challenge is not simply building composable teams, but ensuring that their outputs can be trusted, combined, and audited within a larger ecosystem.

This framework underscores the complexity of scaling. Training, deployment, and coordination of AI agents are already multi-stage processes — from decentralized task definition to compute, inference, and ultimately model marketplaces. Each stage may be governed locally, but coherence requires standards and oversight that extend across the entire pipeline.

The Case for Meta-Governance

Here, the concept of meta-governance becomes indispensable. Governance cannot remain confined to individual clusters or verticals; it must scale across domains. Meta-governance provides the scaffolding for reconciling different local standards, resolving disputes, and ensuring that the agent economy does not fragment as it grows.

This is not about replacing local rules, but about coordinating them, much like constitutional frameworks coordinate diverse local jurisdictions. Without meta-governance, we face a proliferation of agent “city-states.” With it, the Agentic Web can evolve into a federated polity.

DeXe and Multi-Layer Alignment

Meta-governance in the Agentic Web is best understood as a triangle of delegation, expertise, and organization. DeXe operationalizes this triangle through multi-layer structures where expert working groups themselves can function as agents, sometimes as human teams, sometimes as subDAOs, sometimes as hybrids.

Users and other agents can delegate their governance power to these expert clusters. Delegation is reversible and scoped, ensuring flexibility and trust. Experts then translate delegated intent into informed proposals and execution. An automated reward loop ensures that both delegators and experts share in the value created, aligning incentives across layers. Competence attracts capital, and accountability follows expertise without collapsing into centralization.

These expert-agent clusters remain coherent through validator oversight, reputation indexes, and modular governance studios that adapt as clusters evolve. The result is a fractal governance surface: local teams operate autonomously, but align through open incentives and verifiable rules. It’s a toolkit that lets agent clusters specialize, interoperate, and scale while avoiding the gravitational pull of platform centralization.

Toward an Agentic Internet

The authors of Agentic Web argue persuasively that what emerges is not a singular platform but a distributed polity. Its sustainability depends on bridges between clusters and meta-rules that preserve diversity while preventing fragmentation.

With DAO infrastructures like DeXe, such bridges can be built. Standards can be debated, rules enforced, reputations maintained, and oversight distributed. The vision is an Agentic Internet: a public network of autonomous agents coordinated through open governance, auditable standards, and federated institutions.

If composability makes the system powerful, meta-governance is what makes it sustainable. The next internet will not be engineered solely by protocols or platforms, but by the alignment of agents within systems of governance capable of evolving with them.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp

DeXe is launching a new community event focused on user-generated content. The goal is simple: boost the presence of $DEXE across Web3 through creativity.

Memes, AI images, threads, articles, or short videos — any format works as long as it shows what $DEXE, DeXe DAO, or DeXe Protocol are about.

User content has always played a big role in the growth of Web3. DeXe is ready to reward the most active and creative contributors.

Timeline

Starts: October 6, 2025
Ends: October 17, 2025

Post your content on Twitter using $DEXE and #DeXeCreativityCall.
Make sure to tag @DexeNetwork.
After the event ends, submit all your content using the Google Form.

What you can create

This is a content-first event. There are no limits.
Here are some examples of what we accept:

• Memes
• AI-generated images
• Threads and thought posts
• Articles and explainers
• Short videos or edits
• Anything else that tells the story of $DEXE

Rewards

Top 30 participants will receive $DEXE rewards
Winners will be selected by the DeXe Marketing DAO based on originality, consistency, and reach.

Selected entries will be featured in our recap post

How to join

1. Fill out the Google Form: https://forms.gle/8uSoGm4PMEpzhed36
2. Create content about $DEXE.
3. Post it on Twitter with the $DEXE tag and mention @DexeNetwork.

Content review starts on October 17. Rewards will be distributed shortly after.

Keep it creative. Keep it real. DeXe Creativity Call starts now.

Stay tuned

Website | Twitter | Telegram channel | Telegram chat | dApp