The global financial system is moving towards an infrastructure in which artificial intelligence agents execute transactions, manage capital and make payments autonomously. In June 2026, three simultaneous launches—Coinbase, Visa, and Tether—presented initiatives that aim to enable what they describe as an “agentic economy,” although these are implementations still in early phases and with different levels of technical and regulatory maturity.
The trailer comes after a series of recent releases that enable for the first time the execution of transactions by artificial intelligence agents in real environmentsboth on exchange platforms and on payment networks and physical devices.
Coinbase presented “Coinbase for Agents”, a CLI and MCP system that allows you to connect artificial intelligence models to user accounts through API keys with limited permissions. The system enables the execution of operations such as trading, transfers and portfolio management under programmable security conditions, including isolated subaccounts, operation limits and prior validation of transactions.
Visa, for its part, reported that its stablecoin settlement infrastructure already reaches an annualized volume close to 7 billion dollarsand announced explorations to enable payments initiated by AI agents in collaboration with third parties.
The stated goal is to enable automated purchases within digital environments with user-defined limits, while Visa would act as a verification and dispute resolution layer. However, the model introduces questions about liability, transaction reversibility, and control in self-executing scenarios.
On the other hand, Tether led an investment round of up to $1.4 billion in NEURA Roboticswith the aim of integrate self-custody wallets directly into humanoid robots. The initiative includes the integration of custody and payment tools into robotic systems. NEURA Robotics develops humanoids capable of interacting with the environment through advanced sensors and artificial intelligence models, in what it calls “Physical AI”, with industrial and domestic applications.
It is worth noting that, in parallel to these announcements, during 2026 a broader trend was consolidated in the ecosystem: the emergence of wallets designed for artificial intelligence agents. MetaMask, Coinbase, Cobo and Trush Wallet are some of the companies that focused the design of their wallets on operational interfaces for autonomous agents with economic execution capacity.
The challenge of security in autonomous systems
The advancement of these tools responds to a structural change: digital markets operate continuously, while artificial intelligence agents are now capable of executing actions without constant human intervention. However, Autonomy introduces new layers of risk. Configuration errors, excessive permissions or vulnerabilities in systems can lead to loss of funds or unforeseen behavior.
In this context, NVIDIA presented SkillSpector, a pre-audit system that evaluates agent skills before execution. The tool analyzes 64 types of vulnerabilities in 16 categories, including data exfiltration and privilege escalation. Likewise, its system of scoring assigns scores from 0 to 100, automatically blocking any evaluation above 50.
It is important to mention that, according to data from the ecosystem analysis, 26.1% of the “skills” evaluated present vulnerabilities, while 5.2% show high risks or potentially malicious behavior.
For now, the development of autonomous financial agents reopens a central problem: how to guarantee security in systems where execution is automatic and continuous. The challenge is no longer enabling these capabilities, but rather defining what type of control and verification is imposed when economic execution occurs autonomously and persistently within the global financial system.