ChatGPT Pro now connects to 12,000+ banks through Plaid. Read-only, for now, but the message is clear: the fight is over who owns the customer’s financial context, not who ships the best chatbot.
Banks still hold the decisive moat. Behavioral signals, fraud scores, action capability (HELOCs, trades, credit decisions). Plaid gives ChatGPT eyes on the data; banks still hold the hands on the levers. The strategic imperative is to build the context-and-action layer before someone else does.
Three things moved this week. First, Canada’s DGSI/TS 140 open finance spec entered public review, closing September 28. It is the first Canadian technical standard for Open Finance data portability with explicit AI data governance and traceability for ported data. The Consumer-Driven Banking Act mandates a single API standard, 99.5% uptime, 24 months of transaction history, and Bank of Canada supervision. Banks operating in Canada need API mapping and accreditation readiness now, not next quarter.
Second, Know-Your-Agent infrastructure is racing ahead of banking rules. Mastercard Agent Pay is live with 30+ partners. AGENPAY is building a KYA compliance engine. Coinbase’s board has mandated a “financial system for AI.” XRPL agent payments surged 29% in four days to roughly four million transactions. Banks are blocking agent fund access, appropriate for now, but deposit leakage to stablecoin-settled agent flows is the trigger point to watch.
Third, the governance gap is becoming a regulatory risk. 52% of financial services firms are deploying agentic AI, but only 33% have governance safeguards. Only 8% can apply predictive insights to live campaigns. The bottleneck is data infrastructure, not model sophistication. OSFI’s AGILE framework and Guideline E-23 will force closure on a compressed timeline.
The pattern is consistent: the intelligence layer matters more than the chatbot. Canadian regulation is converging in Q4 2026, with RTR, Consumer-Driven Banking, and DGSI/TS 140 all arriving together. Agent identity infrastructure is being built by payment networks and exchanges, not banks. The banks that treat this as a data infrastructure problem, not an LLM problem, will be the ones still in the room when the agent makes the next decision.