Author: Prathik Desai
Compiled by: Chopper, Foresight News
More than a decade ago, I made my first online purchase with a credit card. My mother was very uneasy. Why would you trust a merchant hiding behind a website whom you've never met? What if the action camera never ships? At the time, I couldn't answer her. Now I have the answer.
Back then, the aggregator platform Flipkart (now part of Walmart) stepped in between an ordinary seller and me, providing a layer of trust assurance for the merchant.
Every commercial era expands the boundaries of transactions between strangers, and each era requires new types of intermediaries to foster trust. As David Graeber described in his book "Debt: The First 5000 Years," even before the birth of currency, trade among the Sumerians and Babylonians operated on credit and social relationships. When banks were unwilling to lend to strangers, credit bureaus issued personal credit scores; when credit cards emerged, Visa and Mastercard provided credit guarantees for every swipe; and when the internet gave rise to a vast number of anonymous sellers, eBay, Amazon, and Flipkart managed the trading ecosystem through rating systems.
Then, the machine age arrived.
This article will explain how to fill the trust gap when the total transaction volume of machines is about to surpass that of humans.
Non-Human Buyers
All the trust systems mentioned earlier can only work under the premise that the transacting entities are humans or businesses. If a corporate entity defaults, we can trace it back to specific responsible individuals.
But this logic does not apply to autonomous agents. Today, machines can make purchases according to instructions set by human principals. Meanwhile, AI agents are increasingly generating sub-agents autonomously, issuing procurement tasks to achieve overall goals.
These agents are just lines of anonymous code and software; at first glance, it's often impossible to trace the actual controller behind them.
Agents can now book travel, negotiate prices, settle bills, and even call other software to handle tasks they cannot complete themselves. The entire process is extremely fast, leaving no time for humans to monitor in real-time. In the time it takes you to read this sentence, AI agents have already completed hundreds of transactions. Simultaneously, the scale of funds involved in these transactions continues to grow, making this issue impossible to ignore.
So, if an agent does something that doesn't conform to general trading rules, who should you hold accountable? These agents can change their identities at will. They can adopt a new identity and start over in a matter of seconds. Cryptocurrency is helping us solve this problem.
The Transferable Identity Problem
In January 2026, Ethereum launched the ERC-8004 standard, which can generate a permanent identity identifier on the public blockchain for each agent, while also recording the tasks completed, ratings received, and risk warnings for that agent. Anyone can publicly verify this record without needing permission from a centralized authority.
This identity serves both as a pass and a reputation file. But it has a flaw: the standard issues the agent's identity as a transferable NFT smart contract, known in the crypto world as an ERC-721 token. This transferable nature introduces risks: agent identities can change hands.
If an identity can be sold, and reputation is tied to that identity, then reputation will also transfer, leading to a series of hazards.
For example: someone deploys an agent that operates stably and compliantly for six months, successfully completing tasks, accumulating positive reviews, and building a high credit score. They then sell this identity at a high price to someone else, and the new holder directly inherits this reputation. The original deployer and principal can cash out and exit. If the new holder has malicious intent, they could use the inherited reputation to engage in违规 behavior.
How do we plug this loophole?
The Identity and Trust Tech Stack
First, let's look at the traditional trust systems that agent commerce is poised to replace.
The credit of merchants and online sellers is not certified once and valid forever; their credit is dynamically assessed based on continuous behavior. A few disputes or reports of fraud can restrict their market access and directly damage their reputation. Online aggregation platforms often impose fines on involved merchants, require additional保证金, and have the authority to permanently delist them.
Agents need a similar evaluation mechanism. Currently, most people simplify agent trust to a scoring problem, believing that just measuring metrics is enough. The real difficulty lies in: tightly binding the assessment results to the agent itself, preventing credit from being traded or laundered.
A core prerequisite for an agent to be credible is that its funding source can be traced back to a human principal.
Skyfire, backed by investors like Coinbase, a16z, and Circle, has launched an agent trust tech stack to ensure that the交易对手 is a verifiably identity-authenticated, trustworthy agent. Skyfire issues signed identity tokens for agents, based on cryptographic principles equivalent to the verification mechanisms used by bank websites to prove their authenticity to users.
Skyfire's tokens simultaneously bind three pieces of information: the platform that developed the agent, the agent itself, and the individual or company acting as the principal. When an agent initiates a transaction, the seller can verify the token, and in case of issues, can clearly identify who to hold accountable.
Skyfire also includes a payment layer; this token used for identity verification can also be used for payment settlement.
As agent payments become widespread, the settlement layer will gradually become homogeneous, with narrowing value opportunities. Value will concentrate in the identity and trust layers. Cryptographic "credit bureaus" will build systems to complete agent identity verification, credit背书, and transaction execution.
Several startups are racing to capture value in this赛道 using different approaches.
Skyfire anchors trust to an external principal, while RNWY chooses to attach credit directly to the agent itself. RNWY adopts a bound token (Soulbound Token) scheme, inspired by a reputation system that is already thirty years old.
It cites a perspective from a paper in the Cambridge Knowledge Engineering Review: within an ecosystem, changing identity comes at a relatively high cost. This can prevent agents from easily escaping consequences by switching accounts after违规 behavior.
Soulbound Tokens serve as this cost constraint. Regular ERC-721 NFTs can be freely transferred between different wallets, but Soulbound Tokens issued based on the ERC-5192 standard remove the transfer function. Once minted, the token is permanently bound to the corresponding wallet, much like biometric information is bound to a passport.
Therefore, the only way for an agent to erase a bad record is to abandon the wallet holding that token, along with all the reputation accumulated in the past. This is the违约成本 under the institutional design. This token standard fundamentally increases the cost of abandoning credit.
Currently, RNWY has registered over 230,000 agents based on this mechanism, all through ERC-8004 contracts for identity registration.
If an agent wants to start over from scratch, it must forfeit all its historical records. RNWY deliberately raises the cost of "changing identities and running away." Only by paying this price can one obtain a non-transferable, hard-to-manipulate credible资质. This mechanism also constrains long-term compliant merchants, preventing subsequent违规 behavior.
However, this binding model is a double-edged sword: if you lose the private key corresponding to the Soulbound Token, you lose the agent's identity, credit, and all its long-accumulated behavioral records.
ChainAware takes a different path. It doesn't issue passes or identity credentials. Instead, it reads on-chain behavioral data from wallets and uses it as a basis to generate wallet credit scores. Its model is trained to identify addresses associated with known scams and addresses with clean histories.
Its core logic is not to trace problems after the fact but to predict risks in advance, similar to how a bank's risk control department identifies异常 transactions and freezes cards. The credit score is entirely generated from real behavior; good reputation cannot be bought.
A host of developers are racing to extract value from the trust and identity layer, each with different approaches: Skyfire anchors trust to a principal; RNWY binds credit via non-transferable tokens; ChainAware skips identity issuance and directly scores agents based on wallet behavior.
I believe no single solution can address all scenarios. The启示 from these projects is that for the agent trust layer to capture significant value, it must build a complete tech stack. The底层 must be able to lock in the principal's responsibility; the中层 must ensure that reputation and identity are non-transferable; on top of that, behavioral scoring systems can provide crucial references for merchants and agents, distinguishing between the transacting agent and its背后 principal, and assessing credit based on real, un-purchasable behavior. Each module together forms an identity and trust stack, much like how bank transactions add layers of SMS verification codes, two-factor authentication, and passkeys for security.
Humans spent a long time teaching agents to make payments, retrieve information, and collaborate with each other. But as with all commercial eras, the ultimate direction of the agent economy depends on how we build the trust system.
Throughout the cycles of商业发展, the intermediaries that can solve trust problems always capture the most value. This wave of agent浪潮 will likely be no exception.
