Ripple Treasury has expanded its GSmart platform with policy-controlled artificial intelligence agents for forecasting, liquidity management, risk monitoring, reconciliation, and reporting — while insisting that no transaction moves without a human signature.
The update, announced Sept. 10, adds what the company calls the industry’s first governed AI for enterprise treasury. Specialized AI agents examine treasury data while enforcing controls set by each company, and the architecture deliberately separates the machinery of finance from the interpretation layer that explains it.
Calculations and explanations, kept apart
The core design decision is the split between deterministic engines and the AI layer. Rather than letting a language model calculate financial results and act on them, GSmart keeps all financial math inside deterministic engines. The AI reads company policies, identifies patterns in treasury data, and explains its recommendations — but never produces the numbers those recommendations rest on.
Finance teams retain control over each transaction because the system requires a person to approve any financial action. Ripple Treasury says this gives companies access to more automated analysis without handing final authority to an AI agent — a distinction that matters in a department where a wrong number can move real money.
Each agent is scoped to a defined treasury function — cash forecasting, liquidity management, financial risk, account reconciliation, or reporting — rather than operating as a general-purpose assistant roaming across a company’s financial systems. When an agent detects activity that may breach an internal rule, it alerts the treasury team and identifies the specific policy behind the recommendation, letting users review the relevant clause and supporting data before deciding whether to proceed.
Traceability is the selling point: instead of an unexplained warning or a bare proposed action, GSmart shows which internal control produced the result.
Governance layer and adoption numbers
Knowledge Studio acts as the governance layer, where treasury teams define the policies, permissions, and internal controls that guide how the agents examine financial information. Within Analytics Studio, a feature called Ask GSmart provides a conversational interface over treasury data — users submit questions in everyday language and retrieve answers without building a report or manually cross-checking multiple datasets.
The expanded features are already live across Ripple Treasury’s enterprise customer base rather than being tested through a limited pilot. The company reported that 60 percent of customers eligible for Risk Insights have activated the feature, which searches for unusual exposures and possible policy breaches. Forecast Insights has reached 44 percent adoption; it compares expected cash flow with actual results and surfaces gaps that could affect liquidity planning, flagging potential cash shortages or excess balances for staff to assess.
Reconciliation tools round out the update, helping users find differences between financial records, while reporting functions organize treasury information for review. In every case, the automated analysis sits before the approval stage — software can find an issue and recommend an action, but authorized staff must sign off before anything executes.
The governance gap
Ripple Treasury framed the release around a widening governance gap in enterprise AI. Citing Gartner research, the company said an average Fortune 500 business could operate more than 150,000 AI agents by 2028, yet only 13 percent of organizations believe they currently have the governance needed to manage such agents properly.
“Every CFO is under pressure to embrace AI,” said Renaat Ver Eecke, senior vice president at Ripple Treasury. But financial decisions, he added, also need to be explainable, governed, and compliant — which is exactly what GSmart’s separation of calculation and interpretation is built to address. The model leaves employees with a record of the policy used by an agent before they accept or reject its recommendation, a paper trail designed for auditors and regulators rather than convenience.
From GTreasury to Sibos
Ripple Treasury grew out of GTreasury, the treasury management software company Ripple acquired in 2025 and rebranded. The platform handled more than 13 trillion USD in transaction value during 2025 across over 1,000 customers — a figure covering activity recorded through the former GTreasury business, not XRP settlement volume.
The separation is worth noting: Ripple Treasury serves corporate finance functions that extend well beyond blockchain payments, covering cash management, forecasting, reconciliation, risk oversight, and reporting, while Ripple’s digital-asset and payment businesses run separately. The GSmart expansion layers governed AI onto those existing workflows.
The company plans to demonstrate the expanded system at Sibos 2026, scheduled for Sept. 28 through Oct. 1 in Miami, where banks, payment companies, and corporate treasury teams will gather to discuss AI, payments, and digital finance. Expect the governed-AI pitch — deterministic math, policy-referenced recommendations, mandatory human approval — to be the centerpiece.
For the broader industry, the launch is a data point in an unsettled debate: as autonomous agents multiply across finance departments, the firms building treasury infrastructure are betting that the market will pay a premium for AI that can show its work — and knows when to stop.
Keeping the financial math out of the LLM is the whole ballgame. Every enterprise AI demo I have audited happily hallucinated totals when you pushed it.
every demo hallucinating totals is exactly why this sales pitch writes itself. auditors dont care how smart the agent is, they care that 2+2 always equals 4
so its basically a rules engine with an ai chatbot bolted on top lol. smart move tbh, no treasury team would sign off on autonomous agents anyway
That is exactly why it will sell. CTOs can tell auditors a human signs off on every single transaction. Compliance teams love a clean audit trail.
rules engine with a chatbot bolted on is the correct architecture though. the governed ai branding is for the procurement deck, the deterministic engines are what actually move the money
yep and that is exactly why it will pass procurement. no hallucinated payments possible, worst case the bot explains a transfer poorly
disagree, a bot that explains a transfer wrong is still an audit finding. misreported payments get cited the same as missed ones, procurement wont care that the deterministic ledger was technically right
keeping the financial math in deterministic engines and letting the AI layer only explain is the right split. LLMs doing treasury arithmetic is a disaster waiting for a Friday afternoon
Exactly this. I have watched an LLM confidently mis-sum a twelve row invoice. Keep the arithmetic where it can be re-derived and verified.
The Gartner figure is the real story here. 150,000 agents per Fortune 500 by 2028 and only 13 percent of organizations think their governance is ready. Most treasury teams are nowhere close.
and the gartner bit that scared me is governance readiness at 13 percent. the tooling is arriving two years before the policies are, that gap is where incidents live
Governance readiness at 13 percent while the tooling ships anyway is the real story. That gap gets closed by an incident, not by a policy committee.
agree on the engine split but the human signature thing is doing heavy lifting. wait till some team rubber stamps approvals at 5pm before a long weekend
Quorum thresholds handle most of the 5pm rubber stamp risk. Two approvers above a fixed limit and the long weekend problem shrinks fast, assuming treasury teams actually configure it that way.
quorum thresholds fix the 5pm stamp problem tho. two approvers above a limit and the long weekend rubber stamp becomes someone elses signature problem
the reconciliation agent alone justifies the platform tbh. half the treasury teams i know still run month end recs off a spreadsheet nobody has audited since 2021
60 percent adoption on risk insights already and its live across the whole customer base, not a pilot. whoever ran that rollout knows procurement cycles are the real moat
Ask GSmart answering plain language questions over treasury data is the feature that actually gets used. Analysts will lean on it long before they trust any autonomous agent.
ask gsmart answering questions over treasury data is nice until someone asks why a payment bounced and it invents a reason. the deterministic engine should field that one too
the engine split is smart until the explanation layer describes a number the ledger disagrees with. first audit finding like that writes a fun incident report
first time that happens the incident report will 100% blame the ai for misrepresentation of ledger state. whole new error category invented for procurement
blaming the explanation layer only works until the deterministic engine rejects a valid transfer in front of a client. then the incident report writes itself
an explanation layer that disagrees with the ledger is a config bug, not a mystery. banks already reconcile swift messages against core banking, same discipline applies here
the forecasting agent quietly running cash predictions is the thing analysts will actually keep using. everything else is compliance theater for the rfp deck
a forecasting agent quietly doing cash predictions while a human signs off on every transaction is the least sexy and most sellable thing ripple treasury has shipped in years