Operational AI: Proof, Payoffs, & Pressure Points

Starling launches the UK’s first customer-facing AI scam detector, Wells Fargo scales agentic AI under built-in oversight, and Lloyds formalises accountability by placing AI inside its model risk office. At the same time, the EU delays parts of the AI Act even as supervisors signal accelerated enforcement through existing prudential and conduct regimes. We look at Gemini 3’s leap in enterprise capability, the U.S. Genesis Mission’s push toward national agentic infrastructure, growing chip-supply constraints that threaten AI scaling plans, and China’s shift offshore to bypass accelerator bans. Together, these stories show a sector moving from experimentation to execution, where customer protection, agent governance, compute availability, and geopolitical alignment define the next phase of AI in financial services.

UnitedHealth and BNY Mellon report tangible cost savings and cultural shifts as AI moves deeper into production. OpenAI expands its ecosystem with Agent Kit and ChatGPT Atlas, while Visa and JPMorgan tighten controls amid rising agentic payments and AI-driven fraud. We also explore OpenAI’s Project Mercury and what it signals for automation in finance, and review new global guidance from the Financial Stability Board as regulators race to keep pace with AI adoption.

Mastercard prepares to roll out agentic AI payments with new verification, tokenisation, and governance frameworks, while SWIFT trials federated learning to increase cross-border fraud detection. We also look at OpenAI’s restructuring under Microsoft and what it signals for vendor governance, the rise of consensus tools like Eye2.ai as a new compliance layer, and Stanford’s evidence of early-career job disruption that threatens future FS talent pipelines.

We break down the launch of the new GPT-5 suite, examining its expanded capabilities and improvements, but also the issues that remain. We discuss the challenges of context window size, the complications of an unclear routing system, and how these technical details impact real-world use. Finally, we reflect on how the launch itself was managed, what it signals for the industry, and the broader implications for transparency and AI governance.

Nvidia research team suggest Small LMs for agentic AI, OpenAI releases open-source on-device models, and Stanford study suggests workers want more control over AI. We look at how hyper-personalisation may create challenges for managing customer expectations, and how deepfakes are challenging bank security and verification. We also cover reactions to the White House’s new AI action plan, and how big tech is responding to the EU’s AI Code of Practice.Learn more about how we help banks navigate AI strategy, transformation and governance at www.anordea.com

Lloyds bets on hallucination-free AI through UnlikelyAI and launches a firm-wide Data & AI Summer School. Global AI governance tightens, with BRICS pushing for controls, the Czech Republic banning DeepSeek, and the U.S. Senate enabling state-level regulation, raising fresh financial reporting and compliance risks. We also look at TSMC’s record AI chip revenues as a signal of sustained demand, and how Microsoft and Amazon are driving efficiency and capability through AI at scale.Learn more about how we help banks navigate AI strategy, transformation and governance at www.anordea.com