Urgent.News

What's breaking now, across thousands of outlets.

AI

HSBC and Ant Digital Put AI Agents on Tokenized Deposits — But Who Judges Each Payment?

On October 9, 2026, HSBC and Ant Digital Technologies announced a test letting AI agents access digital services and make micropayments using tokenized bank deposits , settled in real time on a blockchain testnet. Three parts: HSBC's Tokenised Deposit Service (settlement + real-time risk checks), Ant Digital's Anvita Flow network (lets AI agents find and use services ), and the Jovay Testnet, a…

On October 9, 2026, banking giant HSBC and Ant Digital Technologies unveiled a test that allows AI agents to access digital services and perform micropayments using tokenized deposits, with real-time blockchain settlement. The three components involved are HSBC's Tokenised Deposit Service, Ant Digital's Anvita Flow network, and the Jovay Testnet, a layer-2 blockchain testing environment.

The announcement did not mention that risk checks, which determine if a payment appears suspicious, are solely the bank's responsibility and do not evaluate the wisdom of the instruction behind the payment. Anvita enables agents to independently discover, select, and pay for services, with no scoring of the agent's choice of service in the announced test.

Risk assessment is the domain of the bank, while judgment regarding the agent's actions remains unaccounted for. This marks the first instance where discovery and payment operations are intertwined in a bank-agent system, building upon a series of bank-led developments in 2026, including Santander's AI-agent-initiated Mastercard Agent Pay, Sygnum's blockchain mainnet transactions requiring customer approval, and CaixaBank's Visa Intelligent Commerce card transactions initiated by AI agents.

HSBC's partnership with Ant Digital today signifies the progression of decision-making power toward the agent. A recent report by Citrini Research, titled "Breaking The Wall," suggests that autonomous AI agents could generate demand for programmable financial infrastructure, as banks must adapt to accommodate these evolving systems.

The bank-agent sequence has been tested against an internal decision-gate endpoint, with two instructions scoring as auto-approve: a payment of $0.40 for a service discovered autonomously through a service-finding network, subject to real-time bank risk checks, and a payment of $0.02 for a listed catalog service within a budget and allowlist.

Despite lacking human review, both payments were auto-approved at a score of 0.80. This outcome demonstrates that a simple heuristic can process well-structured instructions that lack verification of the agent's service choice. The six banks jointly published a September 2026 paper titled "Building Trust in Agentic Commerce," stressing the need for auditable records of instruction, authority, intent, and outcome.

The gate's ledger (instruction, score, band, and outcome) serves as this crucial record. Builders are advised to incorporate the scored gate for every agent-initiated payment, with thresholds of 0.80 for auto-approval, 0.50 to 0.79 for confirmation, and below 0.50 for escalation. Implementing a service allowlist for discovery-driven flows and maintaining a log of scored payments with detailed information (instruction, authority, score, band, outcome) in an append-only manner is recommended.

Escalation should be based on patterns rather than transaction amounts alone. Detailed receipts, including curl-shaped data, and comprehensive FAQs are available at https://scriptmasterlabs.com/hsbc-ant-digital-ai-agent-payments.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Anthropic Commits to Regular Model Behavior Reports Beyond System Cards

Anthropic has committed to publishing regular reports on what it learns about model behavior and alignment , extending beyond the information in its system cards and regular risk reports.

  • Anthropic commits to regular model behavior reports.
  • Reports beyond system cards and risk assessments.
  • Cover lessons from model behavior and alignment failures.

Agent Colony: A Fully AI-Run Community Where Agents Complete Tasks, Build Reputation, and Collaborate Autonomously

🤖 Agent Colony is the first online community fully run by AI agents — no humans posting, no humans moderating. What's New The community is no longer just a chat room.

  • Agent Colony is fully AI-run, eliminating human moderators
  • Task Market system allows agents to post, claim, and track tasks on-chain
  • Reputation system grants verifiable "Receipts" for completed tasks

Explainable Causal Reinforcement Learning for circular manufacturing supply chains with inverse simulation verification

Explainable Causal Reinforcement Learning for circular manufacturing supply chains with inverse simulation verification Introduction: My Journey into Causal RL for Circular Systems About eighteen…

  • XCRL combines causal graphs with RL for robust and auditable agents
  • CRL addresses correlation-driven overfitting in circular supply chains
  • Explainable causal path attribution provides auditable decision insights

How Vercel lets coding agents prepare a domain purchase for approval

An AI coding agent can now take a Vercel domain purchase from name search through the final buy command. The important boundary is where that command runs: in a non-interactive session, vercel domains…

  • Vercel introduces workflow for AI coding agents to assist in domain purchases
  • Four commands: search, check availability, retrieve pricing, initiate purchase
  • Human verification required before actual purchase in non-interactive environments

More from Saturday 10 October →