BankingNewsAI Daily Brief ·
EMVCo issues framework targeting consumer intent in agentic payments
Banking AI
Financial institutions & fintech technology
EMVCo framework targets consumer intent in agentic payments
For the head of cards at a US bank
Your issuer stack treats a valid credential as enough evidence of consent. That fails when an agent has a credential but exceeds a customer’s limits; build authorization and dispute logic around delegated intent before standards set the rules.
EMVCo is developing a technical framework to carry consumer-authorized intent, agent identity and agentic-payment indicators through card payments. Its proposed Intent Services would let participants register, reference, retrieve and manage authorized intent before, during and after a transaction. A customer could set limits on price, merchant, product, timing, recurring purchases or a budget across transactions, while an agent holds a valid credential. EMVCo says historical intent may be needed for disputes; its task force is collating public feedback and may assess changes to EMV 3-D Secure, tokenization, Secure Remote Commerce and Digital Payment Credential.
→ Action
Cards authorization: Map whether authorization records can retain agent identity, transaction indicators, intent limits and historical evidence for disputes.
Read article → from PYMNTS
Colorado proposes AI rules for lenders and chatbots
For chief compliance officer at a lender serving Colorado residents
Treat this as a control-gap exercise, not a legal watch list. You need to decide whether your credit models and customer bots can meet prescribed explanations, human review and disclosure duties before the comment window closes.
Colorado’s Attorney General filed proposed rules on August 11, 2026, for the Automated Decision-Making Technology Act and Chatbot Safety Act; comments close October 26, and both statutes take effect January 1, 2027. Lenders must give plain-language reasons for adverse decisions through more than one channel, provide always-available information, answer data correction requests within 45 days, and retain documentation for three years. Required human review must use an independent, trained person who can reverse the automated decision. Public-facing humanlike chatbots must disclose that they are software, limit claims of professional authority, and meet age-estimation, minor-protection and crisis-response duties; annual reports begin July 1, 2027. Penalties can reach $20,000 per violation.
→ Action
Compliance: Inventory Colorado-facing credit models and public chatbots; map denial notices, data-correction response, human review, disclosure and record-retention controls.
Read article → from Mondaq
General AI
Large language models & AI infrastructure
NCCoE releases comments on AI identity concept paper
For the CISO of a regional bank
Internal agents cannot inherit an employee’s access and call it control. Your bank needs named agent identities, auditable authentication and narrow permissions before agents touch code, data or production systems.
NIST’s National Cybersecurity Center of Excellence released a summary of comments on its Software and Agentic AI Identity and Authorization concept paper and named DevSecOps as its first implementation use case. The project will demonstrate how AI agents can be identified, authenticated and authorized within the software development lifecycle. More than 600 commenters from industry, government and academia submitted feedback on the concept paper. NIST said feedback pointed to early AI-agent adoption in software development. Additional use cases have yet to be determined and scoped, and NIST will review feedback on a rolling basis.
→ Action
Identity and access management: Map internal AI agents to identities and document authentication and authorization controls.
Read article → from Nist
Anthropic invests $100 million to train 10,000 engineers
For the CIO of a large bank
Model access is no longer enough for vendor selection. Anthropic is building Claude-skilled deployment capacity inside firms such as Morgan Stanley, forcing your bank to decide whether its AI vendor can supply trained people through security review and handover.
Anthropic launched Claude Frontier Academy with a $100 million commitment to train 10,000 Frontier Deployed Engineers by the end of 2027. The first cohorts include engineers from Morgan Stanley, Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey and Novo Nordisk. Participants complete a multi-day, in-person program that covers a simulated enterprise deployment, including use-case selection, security review and handoff. Engineers who pass then lead a real Claude use case at their organization during a 12-week residency with Anthropic support. The first Claude Frontier Deployed Engineer badges are expected in early 2027.
→ Action
Technology procurement: Require AI vendors to document deployment-engineer training, project support and security-review handoff before renewal.
Read article → from Anthropic
Apple to add macOS Full Disk Access controls
For the endpoint-security head at a regional bank
AI agents with Full Disk Access can read files, mail, messages and browsing history. Your bank must decide which agents need that access before Apple adds consent steps that can disrupt deployments without reducing the underlying exposure.
Apple said it will add macOS controls around Full Disk Access because AI agents increase the risks of that level of access. Full Disk Access lets an app access files, mail, messages and browsing history, and users can grant it by changing macOS settings. Apple said some developers use the permission in ways that expose everything on a system without users’ full knowledge and understanding. The company said users who want to grant this access will need to take very explicit action. A correction said the change reflects informed consent, not a new permission limit.
→ Action
Endpoint security: Inventory AI agents with Full Disk Access and remove access not required for approved work.
Read article → from TechCrunch
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime