AI agents can buy crypto, borrow against assets, place trades and pay for services. The financial infrastructure is arriving quickly—and the responsibility that comes with it is growing just as fast.
Blockster’s recent coverage shows the pieces coming together: MoonPay connects AI assistants to payments and crypto, Quote.Trade welcomes autonomous traders, and t54 extends credit to software.
That makes NVIDIA’s latest announcement relevant well beyond the chip industry.
On September 28, the company introduced its Open Agent Safety Platform, designed to enforce boundaries around what agents can access and do. As developers give agents more financial capabilities, NVIDIA is working on controls beneath those applications.
Post from @JensenHuangView the original post on X↗
Jensen’s Starting Point: Limit Access
During his September 28 CNBC appearance, NVIDIA CEO Jensen Huang described the starting principle bluntly:
“Job number one is take away all of its rights,” according to published coverage of the interview.
He was discussing an agent’s access permissions: the files, tools, networks and systems it can use. The approach starts with restricted access, granting what is needed for the assigned task.
For finance, that principle translates into familiar distinctions. Permission to check a balance should not automatically allow a withdrawal. Permission to execute a trade should not mean unlimited borrowing. Access to one account should not open every account a customer holds.
The more work agents handle independently, the more those distinctions need to be enforced by the systems around them.
MoonPay Is Bringing Money Into AI Conversations
MoonPay is one of the clearest examples of how quickly those capabilities are expanding.
As Blockster covered in its reporting on MoonPay’s expansion across major AI assistants, users increasingly have ways to buy crypto without starting inside an exchange app.
MoonPay’s PayBox now supports Claude, ChatGPT and Grok, connecting conversations with trading, purchases and payments.
Post from @moonpayView the original post on X↗
Its Kamino integration extends that activity into Solana lending and borrowing. The assistant prepares the transaction, while PayBox handles permissions and execution.
The appeal is easy to understand: describe what you want to do instead of navigating several financial interfaces. But a conversational request still needs to become a precisely authorized transaction, with clear limits on the money involved.
Agents Are Getting Trading Accounts and Credit
Quote.Trade’s V6 launch brings humans, bots and AI agents onto private execution infrastructure covering more than 1,500 markets.
The company lets agents inspect prices and submit trades without requiring human approval for every order. For developers, that creates room for systems that monitor markets and act continuously.
It also raises practical questions: which assets may an agent trade, how large can a position become, and can it move money out of the account?
Post from @BlocksterComView the original post on X↗
Credit adds another dimension. t54, a company building credit and payment controls for AI agents, offers a model resembling a corporate credit card for software. Its Claw Credit product provides controlled credit lines for services such as data and computing power, with repayment histories helping agents qualify for higher limits.
When Blockster covered Claw Credit, t54 reported more than 135,000 agent applications and $500,000 in credit extended.
Agents are becoming participants in financial systems, with spending authority and obligations that extend beyond answering a question correctly.
Payments and Markets Are Becoming More Automated
The same shift is visible across other recent launches.
GoBTC Pay is testing payments between agents and merchants, including transactions in which one agent pays another in Bitcoin. A research agent, for example, could purchase data needed to complete a task.
Rain has opened prediction-market infrastructure to developers and AI agents, providing machine-readable tools for building applications around market creation, trading and settlement.
Underneath those applications, Google Cloud and Puffer are working on faster Ethereum rollup transactions, with sub-second transaction guarantees intended to support increasingly automated activity.
Together, these developments show a broader change: software is gaining more ways to acquire resources, enter markets and move value without someone supervising every step.
What Happens When an Agent Gets It Wrong?
The risks span several different failures.
An agent might misunderstand a request, rely on misleading information or be manipulated by malicious instructions hidden in material it reads. It might also make a perfectly authorized financial decision that loses money.
Blockster explored that distinction in “Letting AI Agents Trade Is the Easy Part. Trusting Them With Our Money Is Much Harder.” In the contributor article, t54 founder Chandler Fang argued that observing a transaction does not necessarily reveal the intent or authority behind it.
That is what makes financial automation consequential. A mistaken answer can be corrected. An executed trade or completed transfer can leave a financial loss that a correction alone cannot undo.
Spending caps, position limits, approved destinations and human approval thresholds become essential parts of the product.
Post from @BlocksterComView the original post on X↗
NVIDIA Is Building Another Layer of Control
NVIDIA’s platform combines OpenShell, software that enforces operating boundaries, with Sentry, a separate hardware-based monitoring system.
Sentry runs on BlueField-4 processors and is designed to quarantine agents that attempt to cross their boundaries within milliseconds. The separation is central to the design: the agent should not control its own monitoring system.
NVIDIA says more than 100 organizations are working with the platform’s technologies, including Microsoft, Anthropic and Salesforce. Citi and JPMorganChase are also collaborating on shared agent-safety technologies. OpenAI, the company behind ChatGPT, is not named in the announced lineup.
OpenShell software is available now. Sentry forms part of a reference system design, and the announcement does not mean every described capability is already deployed across participating organizations.
The crypto projects discussed here illustrate the need for these controls; they are not being presented as confirmed NVIDIA integrations.
Capability and Control Are Developing Together
NVIDIA’s work does not guarantee profitable trading or eliminate financial risk. Application developers still need to define appropriate limits, verify user authorization and handle mistakes.
But it is a concrete engineering response to a problem that becomes more pressing with every new agent wallet, lending integration and trading venue.
The convenience is already attracting users, from financial tools to everyday assistants such as Meta’s Muse. People want software that can complete tasks with less supervision.
The encouraging development is that companies are also building ways to constrain that independence. As crypto gives agents more access to money, Huang is putting NVIDIA’s engineering resources behind making their authority enforceable.
Reporting by Lidia Yadlos
