BankingNewsAI Daily Brief ·
Jensen Huang tells Nvidia staff in a memo that he had an affair with a suspected Chinese spy
Banking AI
Financial institutions & fintech technology
Pliant raises $80m Series C to open payments to AI agents
For the head of commercial cards at a regional bank
Pliant lets Claude, ChatGPT or a client's own agents reach its card issuing and payments through MCP, and banks already build products on Pliant. Offer agent access with spend limits yourself, or rent it from a rival.
Pliant sells one API system that issues company cards, holds payment accounts and converts currencies. Outside AI agents, such as Claude, ChatGPT or a client's own scripts, can plug in through Model Context Protocol, so software can request a card or send a payment mid-task. Pliant raised $80m, passed $100m in annual recurring revenue and will push into the US and Canada. The simple model: the agent pays wherever it is plugged in. That is imperfect, because a client will not plug in without guardrails, and setting limits is old bank work. The exception: banks already build products on Pliant. Why rent from a rival? Handing an agent a key is easy. Deciding which doors it opens is the product. Your card program needs that second part.
→ Action
Commercial cards product: Document whether clients' AI agents can request cards or payments, with per-agent spend limits, approvals and audit logs.
Full story →Read article → from FinTech Global
General AI
Large language models & AI infrastructure
Jensen Huang's memo to Nvidia admits affair with suspected Chinese spy
For the chief risk officer at a tech-lending bank
Nvidia's board is investigating its CEO over ties to a suspected Chinese government agent. If your data-centre loans assume steady Nvidia supply, recheck covenants and concentration limits before the board finishes, not after.
On Sunday Nvidia's board opened an independent investigation into CEO Jensen Huang's relationship with Alice Huang, an AI executive described as an agent of the Chinese government. Huang's memo to staff admits the affair, apologizes and says he will cooperate fully. Why would an affair touch a data-centre loan? Usually it wouldn't. The simple model has three parts. Nvidia makes the chips. Buyers build on them; Japan, for one, plans to buy 27,500 Rubin chips. Banks finance the buildings. The memo shows no chip, contract or export affected, and the inquiry may clear him. The exception is the China link. National-security reviews can bring export limits, and then a borrower's delivery schedule becomes a policy question. Banks track key-person risk at borrowers. The larger one may sit at a supplier.
→ Action
Credit risk: List loans whose collateral or repayment depends on Nvidia chip deliveries and test them against an export-restriction scenario.
Full story →Read article → from The Verge
OpenAI valuation hits $1.25 trillion in new funding round
For the leveraged-finance head at a US bank
OpenAI asked for $30 billion at $1.5 trillion and closed $11 billion at $1.25 trillion. Before lending to a lab, or to a data-center builder counting on lab contracts, assume the next raise also comes up short.
OpenAI sells slices of itself for cash. It set out to raise $30 billion at a $1.5 trillion valuation. It has closed $11 billion at $1.25 trillion, Reuters reported. Why would a company priced above a trillion dollars come up short on cash? Because the last buyer sets the price, and buyers got nervous. One Nasdaq selloff erased $2.5 trillion. Some investors now value Anthropic's planned IPO near $1.5 trillion, down from roughly $2 trillion, and it may be delayed. The picture is incomplete: Bloomberg says the round could reach $20 billion in the coming weeks. Still, these labs pay for data centers out of the next raise, like a household paying the builder from next month's paycheck. When the paycheck shrinks, the builder's lender finds out last.
→ Action
Credit risk: Re-underwrite exposures to AI labs and data-center builders assuming future equity raises close below target size and valuation.
Full story →Read article → from PYMNTS
US AI safety center staff are reporting to the White House
For a model-risk officer at a regional bank
The government's AI safety center gets $10 million a year, has no permanent leader and stopped publishing its tests. When examiners ask who vetted your frontier model, the answer has to be your own team.
CAISSI is a Commerce Department division that tests AI models for safety, the way NHTSA runs cars with crash test dummies into walls and publishes the results. It gets $10 million a year, about one-eighth of its UK counterpart's budget. Its director was fired in April, his replacement resigned, and it stopped publishing its reports. Its staff informally report to a White House office. So why not wait for Washington? Because nothing is coming out. In September Trump asked AI companies to police themselves, with no stated monitoring. Agents are the hard case: no standard test exists, and checking only whether a flight got booked can miss a large error in the steps. A bank buying a frontier model gets a car with no crash rating. The test is yours.
→ Action
Model risk: Document in-house validation for each frontier model and AI agent, testing every step the agent takes, not only final outputs.
Full story →Read article → from Apnews
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime