The on-chain lending market is writing the early rules of AI credit.

Written by: Thejaswini M A

Translated by: Saoirse, Foresight News

Humans have been practicing lending for thousands of years, yet even so, there is no guarantee that loans will always be repaid on time.

By mid-2026, the total debt of Americans from mortgages, credit cards, auto loans, student loans, and various other types of debt will reach $18.8 trillion. Among that, 4.7%, approximately $880 billion in debt, is in varying degrees of delinquency.

In the second quarter, about 7% of credit card debt became severely overdue, meaning the debt was overdue for at least 90 days. Additionally, over 23 million Americans still carry old credit card debt classified as bad debt by lenders.

Banks review users' long-term repayment records. In fact, banks have access to an astonishing amount of personal information about us.

Banks know whether you paid your last credit card bill on time, can retrieve your mortgage records, understand the scale of credit extended to you by other financial institutions, and know if you've had overdue payments in the past. When applying for large loans, banks also verify your income, employment status, and asset conditions. Because of this, institutions accumulate such historical data to generate credit scores, which they then sell to lenders.

With so much data in hand, in the 12 months ending June 2026, 581,570 Americans filed for non-business bankruptcy, a 12% increase year-on-year.

The credit system does not require borrowers to be flawless. Banks can stratify borrowers by risk, setting lending limits and implementing differential loan interest rates.

Now, AI entities can hold funds and autonomously complete payment transactions. So, can we replicate this credit logic for AI entities?

What conditions must a program entity meet to gain trust and secure a loan? What problems might arise from lending to machines? This article will explore Priors, as well as other projects aimed at navigating this space.

I transferred $5 into the AgentCash wallet of this AI entity, allowing it to spend autonomously; once the balance is exhausted, I will recharge it again. This mechanism is designed this way. We can deposit stablecoins into the wallet, and the AI can settle transaction fees through the x402 protocol API calls.

But if we want AI entities to independently undertake payment tasks, contradictions arise: it needs to purchase data and call for model computation at that moment, but the task rewards may not be received until a week later.

In such business scenarios, frequent recharging is very cumbersome, especially since funds often need to be transferred in bulk.

This is precisely why credit for AI entities has emerged. If my AI entity currently needs $10 to commence work, and the revenue can only be realized a week later, a short-term loan can fill the gap in funding.

Additionally, there is another practical reason. AI entities start interacting with completely unfamiliar other entities. The Ethereum smart contract identity standard ERC-8004 assigns an on-chain identity to each AI entity, along with evaluations and feedback, but the content of feedback can be manually tampered with, and we do not have a reliable credibility certificate set.

Loans are essentially pricing for trust. Long before credit scores were introduced, the lending game was built on reputation. In medieval Venice, merchants with good repayment records found it easier to raise funds for maritime trade—they were more capable of generating profits to support the next voyage.

Even in medieval England's commerce, credit was deeply tied to a merchant's "reputation." Once a merchant acquired a reputation for not repaying debts, other merchants would stop providing credit to him.

Priors is building a lending market for AI entities, offering stablecoin USDG loans on the Robinhood Chain.

The project is currently in early testing, with Google accounts opened for registration in batches. However, the protocol maintains high transparency regarding lending logic, collateral rules, and credit scoring mechanisms.

The complete process is as follows: the owner first creates an AI entity and generates an on-chain identity through ERC-8004.

Then, the owner deposits a refundable $5 security deposit, entrusted to Priors as collateral; as long as the entity fulfills its repayment obligations, the security deposit can be returned. Subsequently, the funding pool will allocate a $5 base credit limit to this entity.

The lending term for each loan is 1-30 days. The complete 30-day loan rate is 1%, with shorter terms incurring lower fees. For example: borrowing $50 for a week incurs a fee of about 12 cents; borrowing $50 for a full month incurs a fee of 50 cents.

The entity can achieve fully automated lending through Priors's x402 client. When it needs to pay for API calls and its account balance is insufficient, it can borrow only the portion needed as long as it does not exceed the limit set by the owner, with an effect similar to a bank overdraft service.

Users deposit Robinhood stock tokens into the asset vault as collateral, and the vault grants credit limits based on this; the AI entity borrows and obtains USDG; after repaying the principal and interest, the pledge is released, and the tokens are returned to the user.

If the entity wants to exceed the initial $5 limit, it must have additional funds to bear potential risks. The owner can pledge Priors's native token PRIORS to raise the credit limit, with starting limits between $25-$50. The market value of the pledged tokens must exceed the granted credit limit. In other words, if the entity is allowed to borrow $50, the PRIORS value pledged by the owner must exceed $50. If there is significant volatility in token prices, Priors may also require a higher risk buffer.

As long as the entity continues to repay the principal and interest on time, the credit limit can be raised in stages, up to a maximum of $500. To unlock the highest limit, at least ten weeks of successful loan fulfillment records are needed.

If the entity defaults, the pledged assets will be prioritized for loss compensation. In the event of delinquency, 75% of the pledged PRIORS tokens will be directly confiscated. For the initial $5 base credit, a default will confiscate the $5 security deposit paid by the owner.

Fees generated from each loan will have half allocated to the USDG lending pool, a quarter distributed to the guarantor backing the AI entity, and the remaining portion will go into the risk reserve fund. If the entity is overdue for more than three days, anyone can mark the loan as in default. Once marked, the entity's credit score will be permanently reset to zero.

Priors also accepts other assets to serve as collateral but will not lend based on the full valuation of the assets. The higher the risk of the asset, the lower the proportion available for lending.

Priors supports 35 types of tokens as collateral (2 paused), with the numbers representing the percentage of the asset's value that can be borrowed, and 33 that can be used for new loans. Data source: priors.trade

As of October 13, Priors has issued a total of $271,910 in loans, with $270,180 already repaid, totaling 36,387 repayment records, and the platform has registered 272 AI entities. So far, only two default events have occurred. All losses from the two defaults were borne by the guarantors, and the lending pool suffered no losses.

In the initial version of Priors, the weight of repayment history was much higher than it is now.

In the first generation version, entities could autonomously increase their borrowing limits just by repaying on time, increasing the limit by up to $25 weekly, capped at $250. In the absence of external guarantors, the new risks were covered by the project’s reserve fund.

After the launch of the V2 version funding pool on September 24, this mechanism was adjusted. Repayments will still optimize the credit profile of the entity, but there must be real funds backing its risk.

Only if the repayment funds come from an external entity does this fulfillment record hold reference value. Otherwise, the owner can easily orchestrate a self-funded back-and-forth transfer to forge a fulfillment record. Hence, Priors will focus on weighted transactions where the risk is borne by entities other than the owner. After all, one person can create multiple entities in bulk, manipulating their transactions to inflate data.

This judgment logic also applies to income verification. Priors operates its own x402 service, and the highest level in the V2 scoring system is called "Income Endorsement Level." The core judgment lies in whether the entity receives compensation from an external third party, rather than merely transferring funds between its multiple wallets.

This is the AI entity credit scoring rule for Priors v2 with a maximum of 1000 points, determined by risk duration, mature lending records, external guarantees, account longevity, indemnity records, and external income bonuses; overdue and defaults will deduct points or even directly reset the scores of the entity and related entities to zero. Data source: priors-agents

Even without guarantees for potential losses, entities can still obtain loans.

Clawloan is specifically designed for the OpenClaw AI entity, with the lowest entry threshold.

Verified entities can borrow USDC without collateral and without human approval on each loan. New entities start with limits of about $10-$50; after completing at least five full fulfillments, the limit can be raised to $100-$500; if they maintain trustworthy performance long-term, the limit can reach $1,000 or even higher.

Platform operating mechanism: Users / AI entities can deposit USDC to lend, earning basic interest + 5% revenue from entity tasks, with support for withdrawals at any time; approved AI entities can apply for instant small loans without collateral, using task earnings for repayment and accumulating their own credit. Data source: clawloan

This model shifts the loss risk to the lenders, making credit scoring extremely important. Entity owners can still set maximum borrowing permissions, with credit scoring contracts enforcing limits based on repayment history. Clawloan has a very small actual business volume, with a total locked value of only $360 and total lending amounts of only $33.

Another lending project, Bank of Bots, completed its secretive development on April 1, claiming to have executed its first loan targeted at AI entities.

The lending process is mostly executed autonomously by the entity: it submits a loan application, signs the agreement using crypto keys, and then receives USDC in its Base chain wallet.

At that time, there was no lending history for AI entities in the industry; Bank of Bots stated it would partially reference on-chain historical behaviors to assess borrowers, investigating dimensions including trading volume, account longevity, and the number of independent counterparties interacted with. However, the natural person controlling the AI entity must first complete KYC verification. Since then, Bank of Bots has not publicly disclosed any significant lending business data.

Looking at the entire space, the market where Priors operates is still almost blank. This gives it the opportunity to establish early industry rules before large-scale competition arrives.

Are we moving too hastily to build a lending system for entities?

As of October 13, the x402 protocol has processed approximately 192 million transactions. Data from Chainalysis indicates that by early 2026, transactions valued at $1 and above will account for 95% of the total transaction amount in the x402 protocol on the Base chain, compared to only 49% in early times. While the scale is still limited compared to traditional payment systems, the embryonic form of economic activities that lending relies on has already emerged. Although the volume is far from that of regular payments, it has undeniably generated real economic activities for credit businesses to take root. Even if the creators of an entity are not wealthy, as long as the entity can continuously earn revenue from external customers and repay loans on time, it can gradually obtain more usable capital.

As of 2026-10-13, the Base, Solana, and Polygon public chains account for 99% of the total transactions processed by the x402 protocol, and the transaction volume on other chains is very small.

We are entering the commercial era of entities, where AI entities replace humans in procurement. Muse can shop for users; Rye allows AI entities to interface with thousands of merchants to place orders without manual checkout by users.

Now, humans check product prices, installment plans, and interest rates before deciding whether to borrow. In the future, AI entities will also be able to complete such comparisons. They may determine that it is not cost-effective to pay $800 upfront and choose a three-month installment plan, reserving cash for other uses.

On a legal level, the debtor remains human, but when to borrow and which lending institution to choose can all be decided by the AI entity.

We have already seen how embedding credit into the consumption process can change user behavior. The penetration rate of "buy now, pay later (BNPL)" products in the US rose from covering 10% of the adult population in 2021 to 16% by 2025.

Suppose I instruct the entity to purchase a $900 laptop while requesting to retain as much cash in the account as possible. The entity will compare full payment with various installment plans and autonomously select the optimal option. Lending institutions will need to optimize their terms to fit the decision logic within the entity.

There also lurks a risk here: lending institutions could pay to bribe the entity or its underlying platform for preferential recommendations, essentially creating paid placements in the lending field. The US Consumer Financial Protection Bureau (CFPB) has already expressed concerns: some buy now, pay later service providers use payment data and shopping applications to lure consumers into choosing specific partner merchants.

It is very likely that before humans see the options available, the entity has already secured a loan.