Loading today’s stories
Invest1 distinct publisher3 min readPublished
PayBox stays non-custodial, so nobody else is holding the money; what is new is that one sentence can open a leveraged position on a $1.3bn protocol, and the plugin is barred in the US, UK, EU and Australia.
The Investor · Invest desk
invest
Pump.fun's $12m week puts token issuance above every lending market in crypto1 distinct publisher
invest
Scalable Capital puts ChatGPT, Claude and Grok inside the European order ticket2 distinct publishers
invest
The agent payment rail cleared 8.7 million transfers last week, and $367,9501 distinct publisher
invest
A Solana DEX halted trading and says the loss stopped at its treasury. Nobody can check1 distinct publisher
Compiled by The InvestorSomething wrong?How this is made
Start with the utilization, because it frames everything else: roughly $1bn of active loans against about $1.3bn of total value locked [6] puts Kamino's book near 77 percent drawn, with something like $300m of supplied assets not yet borrowed against [15]. New borrow demand arriving through a chat window therefore lands on a venue that is already mostly lent out.
The term worth rereading is the authorization one. A user can require a passkey confirmation on every transaction, or grant the agent standing permission to act by itself inside limits set in advance [7]. Those are two products sharing one name. In the first, the model is order entry with a spell-checker; in the second, it exercises discretion over a position whose value moves after it is opened, and the announcement itself concedes that managing a leveraged loan requires monitoring, collateral awareness, and the understanding that a single sentence can open something needing ongoing attention [12]. Nothing described adds that monitoring; what is added is granularity about who presses the button.
Custody, the usual first question, is answered and therefore uninteresting here. PayBox is non-custodial, control stays with the account holder [4], so the failure mode is not an operator absconding but a poorly specified instruction, which MoonPay names as a live risk [14], executed against liquidation rules that remain Kamino's [8], where liquidators repay the debt and seize part of the collateral [9]. The interface has no ability to soften a margin call it does not own.
This is probably wrong, but the exclusion list reads more like a legal map than a rollout plan: not the United States, not the United Kingdom, not the European Union, not Australia [10], which is a fair approximation of the world where selling retail credit and margin carries a duty to assess whether the buyer should be holding it. The counter-reading is in the same sentence of the source, which allows that access can differ by asset and jurisdiction [10], and MoonPay has published no timeline for going wider [11]. If the plugin appears in the EU or UK with no appropriateness gate attached, the geofence was about listings, not conduct, and my read fails.
Worth naming what MoonPay is not doing with this release. It is not underwriting, not setting rates, not inventing a credit product [3]; the markets, the terms and the liquidation mechanics are Kamino's [8], and the spend goes into a connector at paybox.sh that links a vault to ChatGPT or Claude [16]. That is a distribution bet, consistent with a vault that already did payments, swaps, bridging and Aave yield before credit [5].
It can run a few ways. Passkey mode dominates, every action is a user click, and the suitability question never attaches. Or standing limits become the default, an agent opens a position that later liquidates, and the chat transcript becomes the record of what was recommended and by whom. Or it stays outside the four blocs indefinitely, a small offshore feature, and the precedent that matters is only that agentic crypto moved past the micropayment rails that protocols such as x402 have been processing in volume [13] into collateralised borrowing [18]. MoonPay's own framing calls this an early test of whether conversational tools can carry that complexity without hiding the risk [17]. Early tests set defaults.
Ranked by verification strength, evidence, and original report placement.
MoonPay expanded its AI-powered PayBox vault so eligible users can interact with Kamino, one of Solana's largest lending protocols, through ordinary conversation in ChatGPT or Claude.
The update does not invent a new credit product; it routes existing Kamino markets through a natural-language interface, so supplying assets for yield or borrowing USDC against crypto collateral can be requested in plain English rather than through a DeFi dashboard.
PayBox went live in July as a non-custodial vault that connects wallets to large language models; the user describes an intent, the assistant prepares the on-chain action, and control stays with the account holder.
From the start PayBox supported payments, token swaps, bridging, and yield via Aave.
DefiLlama data cited across reports put Kamino's total value locked near $1.3 billion and active loans around $1 billion.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 29, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One outlet, one announcement
Every load-carrying detail — the launch date, the two authorization modes, the excluded jurisdictions, the absence of a rollout schedule — reaches us through Crowdfund Insider relaying MoonPay's own announcement. The two numbers that give the story its scale are attributed to DefiLlama via 'reports' that are never named, and nobody outside the companies is quoted. The one claim a reader can independently verify is the connector at paybox.sh.
Shipped, and fenced off from most of the market
Shipping is real: the plugin is live and reachable, and it plugs into a venue with about $1bn of loans already outstanding. Usage of the new path is a blank — no user count, no volume, no number of positions opened by prompt — and the four jurisdictions withheld cover most of the regulated retail world. What we can score is availability, not uptake.
Reach of the framing outruns the map
Modest, and mostly not the reporting's fault. 'ChatGPT and Claude manage Solana lending' is a bigger idea than a plugin unavailable in the US, UK, EU and Australia with no disclosed users, and calling the integration an early test of whether chat can handle leverage assumes an outcome nobody can see yet. Against that, Crowdfund Insider keeps puncturing its own premise — the rates, markets and liquidation rules are unchanged, only the instruction method is new — which is what keeps this near alignment rather than well past it.
A launch told in the launcher's terms
The story originates where the product does. MoonPay set the date, the feature description and the authorization framing; Kamino gains routed deposits and borrowing demand; the trade outlet carrying it serves an audience that reads launches. There is no critic, no user, no regulator and no competing integration in the frame, and the parts a company would least like to volunteer — fees, liability for a delegated agent's trade, the reason four jurisdictions are excluded — are the parts absent.
Coherent, uncorroborated
We are fairly confident about what was announced and where it cannot be used — those details are specific, falsifiable and unlikely to be misreported. We are not confident about anything downstream: how much money moves through a chat prompt, whether delegated limits hold up in a fast drawdown, or whether Kamino's cited scale is current. One source, no second reading, and the only forward-looking claim in the story is explicitly untested.