Treasury here means the treasury function of a company or a bank: cash, liquidity, payments and the risks around them. The US Department of the Treasury's own AI documents, including its March 2024 report on AI-specific cybersecurity risks, are on the Treasury page of the regulation tracker.
What are the AI benefits in treasury risk detection?
Four, each on a bank's own record. First, a bad payment is caught before it leaves: Citi's Payment Outlier Detection, launched in 90 countries in June 2019, profiles each client's payment patterns with statistical machine learning and alerts in real time before an outlier payment reaches the beneficiary, leaving the client to approve or reject it. Second, more signals than a rule can read: U.S. Bank says AI-driven systems learn from behavior, transaction patterns, identity signals and network relationships, so they detect account takeover even when credentials appear valid and expose mule networks. Third, speed against new schemes: First Horizon's CFO Hope Dmuchowski said in February 2026 that AI to search client transactions or logins to the treasury management system can be built in "a day or two" rather than months of new code. Fourth, intraday liquidity risk seen early: Societe Generale's Intraday Liquidity Steering, launched in 2023, forecasts the bank's euro account balances in real time, and the bank reports a 10% cut in euro intraday liquidity usage and related reserve costs. None of the four publishes a figure for fraud losses avoided, and the non-generative models behind them fall under the SR 26-2 model risk guidance.
The parent explainer, AI in bank treasury management, covers the client platforms, predictive analytics, liquidity management and generative AI in treasury portals; AI fraud detection in banking covers the fraud models in depth, and model risk management the validation rules they run under.
SOURCES: Citi: Payment Outlier Detection launches in 90 countries (2019) ↗ · U.S. Bank: How AI can help treasurers stay ahead of fraud ↗ · First Horizon at Bank of America conference, CFO Hope Dmuchowski (transcript, Feb 11, 2026) ↗ · Societe Generale: Applying data and AI, flagship use cases (Intraday Liquidity Steering) ↗ · SR 26-2, Supervisory guidance on model risk management (Apr 17, 2026) ↗
What are the benefits of AI in treasury risk detection?
The banks that run it describe four benefits, each on their own record. The first is catching a bad payment before it leaves: Citi's Payment Outlier Detection, launched in 90 countries in June 2019, replaces rules-based logic with statistical machine learning that profiles each client's payment patterns and sends a real-time alert before an outlier payment reaches the beneficiary, leaving the client to approve or reject it. The second is reading more signals than a rule can: U.S. Bank says AI-driven systems learn from behavior, transaction patterns, identity signals and network relationships, so they can detect account takeover even when credentials appear valid and expose mule networks. The third is speed against new schemes: First Horizon's CFO Hope Dmuchowski said in February 2026 that when a new fraud scheme appears the bank can build AI quickly to search client transactions or logins to its treasury management system, needing "a day or two" rather than months of new code. The fourth is seeing liquidity risk within the day: Societe Generale's Intraday Liquidity Steering, launched in 2023, forecasts its euro account balances in real time so treasury can smooth payments, and the bank reports a 10% cut in euro intraday liquidity usage and related reserve costs. None of the four publishes a figure for fraud losses avoided.
| Benefit | Bank and tool | What the bank reports |
|---|---|---|
| Stops an outlier payment before it is sent | Citi Payment Outlier Detection (Treasury and Trade Solutions) | Machine learning builds a profile of each client's payment patterns and alerts in real time before an outlier payment goes to the beneficiary; the client approves or rejects it in CitiDirect BE or CitiConnect. Launched in 90 countries on June 26, 2019. |
| Reads behavior, identity and network signals | U.S. Bank treasury fraud defence | Anomaly detection, behavioral and identity intelligence and network analytics: detects account takeover even when credentials appear valid and exposes mule networks or coordinated activity. |
| Answers a new fraud scheme in days | First Horizon | The CFO said in February 2026 that AI searching client transactions or logins to the treasury management system can be built in "a day or two" when a new scheme appears, not in months of developer testing. |
| Sees intraday liquidity risk early | Societe Generale Intraday Liquidity Steering | Launched in 2023: a machine-learning model forecasts the bank's euro account balances within the day so treasury can smooth payments; the bank reports a 10% reduction in euro intraday liquidity usage and related reserve costs. |
SOURCES: Citi: Payment Outlier Detection launches in 90 countries (2019) ↗ · U.S. Bank: How AI can help treasurers stay ahead of fraud ↗ · First Horizon at Bank of America conference, CFO Hope Dmuchowski (transcript, Feb 11, 2026) ↗ · Societe Generale: Applying data and AI, flagship use cases (Intraday Liquidity Steering) ↗
Machine learning systems for treasury optimization
Seven systems on the record, three jobs. Forecasting cash so it can be positioned and invested: Bank of America's CashPro Forecasting (more than 3,000 companies saved over 250,000 hours in 2025), J.P. Morgan's Cash Flow Intelligence (Amtrak live from May 2023; Domino's cut weekly manual data clean-up by up to 90%) and U.S. Bank's Liquidity Manager with Kyriba (launched November 18, 2025, with 'Cash AI' forecasting, scenario planning, positioning and pooling). Screening payments before release: Citi's Payment Outlier Detection (90 countries since June 2019) and U.S. Bank's treasury fraud defence. Steering the bank's own intraday liquidity: Societe Generale's Intraday Liquidity Steering (2023, a 10% cut in euro intraday liquidity usage). Feeding the models: Huntington's connectivity platform, more than 500 interfaces carrying more than 10 million transaction events a day. Each figure is the bank's own; none has published forecast accuracy.
| System | Bank | What the model optimizes | What the bank reports | Since |
|---|---|---|---|---|
| CashPro Forecasting | Bank of America | Cash forecasts for corporate clients, built and shared in forecasting workspaces inside CashPro, the bank's treasury, trade and cross-border payments platform | More than 3,000 companies saved over 250,000 hours in 2025; workspaces created in Q2 2025 rose 113% above the norm as clients gauged tariffs and interest rates | 2025 figures (release of December 11, 2025) |
| Cash Flow Intelligence | JPMorgan Chase | AI and machine-learning cash forecasting and categorisation in the J.P. Morgan Access platform; Amtrak used the forecasts to invest idle funds | Amtrak live in May 2023; Domino's cut weekly manual data clean-up by up to 90% | Beta October 2022; Amtrak May 2023 |
| Liquidity Manager (with Kyriba) | U.S. Bancorp | 'Cash AI' forecasting from historical cash flows, scenario planning, cash positioning and pooling inside the SinglePoint portal | Launched November 18, 2025; no usage figures published | November 18, 2025 |
| Payment Outlier Detection | Citigroup | Statistical machine learning builds a payment profile for each client and flags payments that do not fit it for review before they are sent | Launched in 90 countries in 2019 after a pilot with 20 clients | June 26, 2019 |
| Treasury fraud defence | U.S. Bancorp | Anomaly detection, behavioral and identity intelligence and network analytics on treasury payments | Detects account takeover even when credentials appear valid and exposes mule networks or coordinated activity | Undated insight page |
| Intraday Liquidity Steering | Societe Generale | A machine-learning model forecasts the bank's own euro account balances within the day so treasury can smooth payments | A 10% reduction in euro intraday liquidity usage and related reserve costs | 2023 |
| Treasury Management Connectivity Ecosystem | Huntington | API-first connectivity to clients' ERP and finance tools with 'AI-enabled transactional data' for predictive forecasting and automated scenario planning | More than 500 interfaces; more than 10 million transaction events a day | October 23, 2025 |
SOURCES: Bank of America newsroom: CashPro Forecasting (Dec 11, 2025) ↗ · J.P. Morgan: Amtrak enhances cash flow forecasting (Cash Flow Intelligence) ↗ · J.P. Morgan: Domino's unlocks efficiency with Cash Flow Intelligence ↗ · U.S. Bank: U.S. Bank launches AI-driven cash forecasting tool (Nov 18, 2025) ↗ · Citi: Payment Outlier Detection launches in 90 countries (2019) ↗ · U.S. Bank: How AI can help treasurers stay ahead of fraud ↗ · Societe Generale: Applying data and AI, flagship use cases (Intraday Liquidity Steering) ↗ · Huntington: Treasury Management Connectivity Ecosystem (Oct 23, 2025) ↗
Which treasury platforms carry these models?
Every treasury AI platform on the record on this site, from the bank and fintech pages where each row is sourced; the status and date are the profile's.
| Platform | Offered by | What the AI does | Status | Latest source |
|---|---|---|---|---|
| CashPro Forecasting | BofA Bank | AI cash forecasting for corporate clients; 3,000 companies, 250,000 hours saved in 2025. | In production | Dec 11, 2025 |
| CashPro Chat | BofA Bank | AI-driven chat used by nearly 70% of corporate clients for account information, transaction tracking and service resolution, with a 43% containment rate in Q3 2025; the Q2 2026 earnings release counts 40,100 CashPro Chat interactions supported by Erica technology. | In production | Jul 14, 2026 |
| SinglePoint Liquidity Manager (with Kyriba) | U.S. Bank Bank | AI forecasting and scenario planning for corporate clients, with Kyriba. | In production | Nov 18, 2025 |
| Citi Treasury and Trade Solutions (Feedzai) | Citi Bank | Feedzai quotes Manish Kohli, global head of payments and receivables at Citi Treasury and Trade Solutions, on Citi's 'strategic partnership with Feedzai', which he says helps it scale so clients can make payments 'securely, efficiently and without friction, across the globe'; Feedzai's 2021 funding release also named Citibank among users of its risk-management software. | In production | Oct 11, 2024 |
| Wells Fargo payments AI | Wells Fargo Bank | A tool that summarises banker-client calls beyond a CRM dump; payments AI turns balance, receivable, payable, FX and liquidity data into cash-flow forecasts and risk flags. 26 of 335 experiments live by May 2026. | In production | May 5, 2026 |
| KeyTotal AR (with Versapay) | KeyBank Bank | Machine learning for invoice-to-cash for middle-market clients, with Versapay. | In production | Jul 7, 2025 |
| Treasury Management Connectivity Ecosystem | Huntington Bank | Intelligent automation across 500-plus interfaces for business clients. | In production | Oct 23, 2025 |
| HSBCnio | HSBC Bank | Combines transaction banking with AI interfaces, including MCP-based data queries and a planned 'Ask HSBC' assistant. | Announced | Sep 29, 2026 |
| RegionsClientIQ | Regions Bank | A machine-learning data product that prioritises client opportunities and flags early credit and attrition risk for commercial and treasury bankers; a gen-AI copilot added in Q1 2026. | In production | Aug 7, 2026 |
| Mercury Command | Mercury Fintech | Natural-language assistant that stages categorisation, letters, analysis, payments and invoices for approval; the model proposes, the product enforces, the user authorises. | In production | Jul 16, 2026 |
| Ramp Accounts Receivable agents | Ramp Fintech | Three AI agents handle invoice creation, collections follow-up and cash application/payment matching in Ramp's Accounts Receivable product; Finance reviews and sends every collections email. Ramp reports a median 36-hour time to payment among early customers. | In production | Sep 22, 2026 |
Ranked by the bank records: which large banks lead in treasury, cash management and liquidity AI.
What is an AI treasury risk assessment?
It means two different things. For a corporate treasurer, it is the use of AI to assess treasury risks: flagging a payment that does not match a company's usual pattern, forecasting cash to spot a shortfall early, or running scenarios for a rate or tariff move. Clients ask their banks for it: in PNC's October 2023 survey of more than 120 finance professionals, 23% wanted AI-driven risk assessments from their bank and 45% predictive analytics and forecasting. For the bank, it also means assessing the risk of the AI itself. Under the US model risk guidance a model's risk reflects its inherent risk and its materiality, and models used 'to help meet regulatory requirements or manage a banking organization's financial risk exposures are generally considered to be of greater risk', a description that fits a bank's liquidity and interest-rate models.
| Treasury risk | What the AI does | Where banks do it | What still needs a person |
|---|---|---|---|
| Payment fraud and errors | Learns each client's payment pattern and flags outliers before release; behavioural, device and network signals for account takeover and mule networks. | Citi Payment Outlier Detection; U.S. Bank real-time fraud detection and network analytics. | Approving or rejecting the flagged payment; Pinnacle tells clients authorisation and verification 'still require people and strong controls'. |
| Cash shortfall and liquidity | Forecasts daily cash positions from historical flows and adapts as conditions change. | U.S. Bank Liquidity Manager ('Cash AI'); J.P. Morgan Cash Flow Intelligence; Bank of America CashPro Forecasting. | Funding and investment decisions; Amtrak used the forecasts to invest idle funds. |
| Market and rate moves | Scenario planning on the forecast: what a tariff or rate change does to cash. | Bank of America clients used CashPro Forecasting to gauge tariffs and interest rates in 2025. | Choosing the scenarios and acting on them. |
| AI-enabled social engineering | Detecting deepfake and AI-generated impersonation in payment requests. | U.S. Bank names emerging-threat detection; Pinnacle warns clients 'Fraudsters have access to AI, too.' | Call-back verification on a known number for any change to payment instructions. |
| Data quality | Consolidating and reconciling data across banks and systems so the models see clean inputs. | Huntington's connectivity platform; U.S. Bank multi-bank reporting. | Deciding which source is authoritative. |
| The AI tool's own risk | Validation of the vendor or in-house model: conceptual soundness, outcomes analysis, ongoing monitoring. | SR 26-2, for non-generative models; the agencies say it is most relevant to banks with more than $30 billion in assets. | Effective challenge by people with the expertise, independence and standing to force changes. |
SOURCES: PNC Treasury Management: AI in action, survey of finance professionals (Oct 2023) ↗ · Citi: Payment Outlier Detection launches in 90 countries (2019) ↗ · U.S. Bank: How AI can help treasurers stay ahead of fraud ↗ · U.S. Bank: U.S. Bank launches AI-driven cash forecasting tool (Nov 18, 2025) ↗ · J.P. Morgan: Amtrak enhances cash flow forecasting (Cash Flow Intelligence) ↗ · Bank of America newsroom: CashPro Forecasting (Dec 11, 2025) ↗ · Huntington: Treasury Management Connectivity Ecosystem (Oct 23, 2025) ↗ · Pinnacle: AI and Treasury Management: More Opportunity, More Risk ↗ · SR 26-2, Supervisory guidance on model risk management (Apr 17, 2026) ↗
Which rules do treasury risk-detection models run under?
SR 26-2, OCC Bulletin 2026-13 and FDIC FIL-15-2026, in force since April 17, 2026, cover traditional statistical models and non-generative, non-agentic AI, so a machine-learning payment-screening or liquidity model is in scope and, because it manages a bank's financial risk exposures, is treated as generally of greater risk. A vendor model such as Kyriba's inside U.S. Bank's Liquidity Manager is a third-party relationship under the 2023 interagency guidance. The US Treasury's own view of AI risk in the sector, including the AI-enabled cyber and fraud threats that treasury payment screening is meant to catch, is in its March 2024 report on managing AI-specific cybersecurity risks and its December 2024 report on AI in financial services.
Is this page about the US Treasury?
No. Treasury here means the treasury function of a company or a bank: cash, liquidity, payments and the risks around them. The US Department of the Treasury's own AI work, including its March 2024 report on AI-specific cybersecurity risks in financial services, its December 2024 report on AI in financial services and the 2026 Financial Services AI Risk Management Framework, is covered on the Treasury pages of the regulation tracker, linked below.
Which bank first used machine learning to screen treasury payments?
On the records here, Citi: its Payment Outlier Detection, launched in 90 countries on June 26, 2019 after a pilot with 20 clients, replaced rules-based logic with statistical machine learning that profiles each client's payment patterns and alerts in real time before an outlier payment reaches the beneficiary.
Do treasury risk-detection models fall under SR 26-2?
The non-generative ones do. SR 26-2, OCC Bulletin 2026-13 and FDIC FIL-15-2026, in force since April 17, 2026, cover traditional statistical models and non-generative, non-agentic AI, so a machine-learning payment-screening or liquidity model is in scope, and because such models manage a bank's financial risk exposures the guidance treats them as generally of greater risk. Generative and agentic AI are outside its scope.
Are the benefits independently verified?
No. Every figure on this page is the bank's own, as published in its release, insight page, conference transcript or use-case document, and none of the banks publishes a figure for fraud losses avoided or forecast accuracy. Where a figure is quoted it is dated and attributed; where nothing is published the table says so.
Which machine learning systems do banks run for treasury optimization?
Seven on the record: Bank of America's CashPro Forecasting, J.P. Morgan's Cash Flow Intelligence and U.S. Bank's Liquidity Manager for cash forecasting and positioning; Citi's Payment Outlier Detection and U.S. Bank's treasury fraud defence for payment screening; Societe Generale's Intraday Liquidity Steering for the bank's own intraday liquidity; and Huntington's connectivity platform that feeds the models. Each is dated and sourced in the table above.
When a bank ships treasury AI, or a supervisor moves on it, you'll read it the next morning.
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning