BankingNewsAI Daily Brief ·
HSBC and Ant Digital test AI agent payments using tokenized deposits
Banking AI
Financial institutions & fintech technology
HSBC and Ant Digital test AI-agent payments using tokenized deposits
For the payments head at a regional bank
Treat agent-initiated deposit payments as an authorization and liability design problem now, not a card-only pilot. Your bank must define agent identity, customer permissions, transaction caps and dispute responsibility before it connects money rails.
HSBC and Ant Digital Technologies tested AI-agent micropayments using tokenized bank deposits, with real-time settlement on a blockchain testnet. The test combined HSBC’s Tokenised Deposit Service, Ant Digital’s Anvita Flow network and the Jovay Testnet layer-2 environment. HSBC provided settlement and real-time risk checks, while Anvita Flow let an AI agent find a digital service, access it and complete a payment. The companies said the test only verified technical capabilities and was not a commercial launch or live customer offering. Micropayments typically mean transactions below $2.
Read article → from Tradingview
PLACE acquires Ardley to build mortgage operating system
For the president of a mortgage bank
PLACE is building a suite that ties portfolio intelligence to origination and fulfillment. Decide whether integrated recapture gains outweigh the loss of control that comes with putting origination, servicing and retention tools under one vendor.
PLACE agreed to acquire Ardley, whose software sends pre-structured loan offers to mortgage-company and bank customers. Ardley helps lenders and servicers use existing loan portfolios for refinance, purchase, second-mortgage and line-of-credit opportunities. The deal adds to PLACE acquisitions of Maxwell, Remine, Envoy Mortgage and divisions from Radian Group. PLACE acquired Maxwell; Maxwell serves more than 400 financial institutions and facilitates more than $130 billion in annual mortgage transactions. Ardley and Maxwell each touched nearly one in 10 U.S. mortgage originations in 2025, according to PLACE.
→ Action
Mortgage technology: Map recapture, point-of-sale, origination and servicing dependencies before selecting an integrated platform.
Read article → from PR Newswire
General AI
Large language models & AI infrastructure
UK ICO secures AI privacy changes from 10 developers
For the chief privacy officer at a US bank
UK privacy expectations now reach foundation-model vendors. Require proof on how personal information is collected and used for training, rights and deletion handling, and agent data use; a general vendor privacy policy is no longer enough for cross-border deployment.
The UK Information Commissioner’s Office said Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI made or committed to make data-protection changes. The ICO closed a supervisory programme that began in 2025 and assessed firms for potential non-compliance with UK data protection law, UK market share and use of higher-risk training datasets. Commitments cover clearer explanations of how personal information is collected and used for model training, plus stronger mechanisms for access, rectification, objection and erasure rights. The ICO also published positions on special-category data, whether trained models may contain personal data, and data-protection risks from AI agents.
→ Action
Privacy: Require AI vendors to document training-data sources, data-rights and erasure handling, and agent data flows before deployment.
Read article → from Tech-insider
Postman runs Agent Mode on Bedrock for 40 million developers
For the CTO at a regional bank
Your agent program needs architecture controls before broader release. Model choice and prompt tests will not stop a large tool catalog, UI state, excess context, or weak data-retention settings from creating operational risk.
Postman runs Agent Mode for 40 million developers on Amazon Bedrock, with the agent working across API testing, documentation, discovery, and implementation. Bedrock lets Postman handle variable, latency-sensitive global demand without running its own model-serving infrastructure, while selecting models and controlling throughput, geographic processing, and cost. Postman found tool-selection errors rose when the visible toolset exceeded about 40 tools; its root agent now narrows more than 170 tools to about 15 for each task and passes them to a context-isolated sub-agent. Users must approve actions that change application state, and Postman uses Bedrock Guardrails to redact personally identifiable information before it reaches the LLM.
→ Action
AI platform: Test task-based tool scoping and require approval for agent actions that change application state.
Read article → from Amazon
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime