Autonomous AI agents have become increasingly proficient at searching and recommending shopping options, but most stop short of completing transactions. Fetch.ai said Thursday that it will roll out a payment system in 2026 designed to let AI agents execute purchases and deposits on a user’s behalf.

The new feature set to launch in January addresses one of the largest barriers to wider adoption of agentic AI. While consumer-facing systems can suggest flights, hotels, or services, virtually all agentic systems still require real-time human approval before handling payments because of security, liability, and regulatory risks.

Key Takeaways

• Fetch.ai plans to enable AI agents to complete payments autonomously starting in 2026

• The system aims to remove the need for constant human approval in routine transactions

• Visa-backed infrastructure and identity controls are central to the design

Fetch.ai founder and CEO Humayun Sheikh said the obstacle was less about access to payment rails and more about how AI systems are designed.

“We’ve been working on it for at least five years, and the reason is because we’re going to see a transition from the web-based economy to an AI-first economy,” Sheikh mentioned: “And the only way to do that is where AI agents can communicate with each other and transact with each other.”

From Recommendations to Transactions

The new payment functionality is hosted on Fetch.ai’s ASI:ONE platform and aims to allow agents to book services, place orders, and send payments even when a user is offline. The company said it is working with established financial providers instead of building proprietary infrastructure.

“All of these rails matter, but having them isn’t enough if a system can only do one thing at a time internally,” he said. “That’s why ChatGPT hasn’t managed to do this. It doesn’t have a system that allows you to actually buy something from someone.”

Security, Identity, and Payments Infrastructure

“This is a Visa system,” Sheikh said. “We’re connecting it to the agent and making sure there’s trust building, a layer of security, and a KYC element built in.” He added that while the technology is ready, the rollout was delayed to January to allow additional checks by Visa. Mastercard support, he said, is expected to follow.

To address security concerns, Fetch.ai said the system relies on single-use payment credentials rather than permanent card numbers. Transactions are authorized through Visa, which issues temporary credentials limited to specific amounts and purposes.

The platform supports both traditional card payments and on-chain transactions using USDC or Fetch.ai’s FET, the native token. The company said it has avoided direct bank transfers, citing the additional regulatory requirements associated with that approach.

Fetch.ai has also built an identity layer into the system, requiring agents to act on behalf of identifiable users or businesses rather than operating anonymously. The design comes as retailers and platforms push back against automated shopping tools that mimic human behavior.

Last month, Amazon sent a cease-and-desist letter to Perplexity, alleging that its Comet tool was disguising bots as human shoppers. Sheikh said Fetch.ai’s agents are designed to operate transparently, with persistent identities tied to specific users.
“When an agent isn’t live, for example if it’s hosted on a local machine that goes offline, it has a mailbox,” Sheikh said. “When it comes back online, it checks messages, downloads them, interprets them, and takes action.”

The rollout also comes amid changes inside the Artificial Superintelligence Alliance, a group Fetch.ai formed in 2024 with SingularityNET and Ocean Protocol. In October 2025, Ocean withdrew from the ASI Alliance, citing disputes over treasury control and the closure of its token bridge.

Despite the turbulence, Fetch.ai said it remains focused on an owner-operated model that allows users and businesses to host and manage their own agents rather than relying on centralized platforms.

“That architecture exists for people who want to run their own agents,” Sheikh said. “We give them the facility to own and operate them, with each agent representing a known user.”