Ripple rolls out governed AI agents for corporate treasury, with a mandatory human-in-the-loop check
The GSmart platform automates liquidity rebalancing, cash forecasting and fraud monitoring, but requires manual approval before any agent-proposed action involving money actually executes.
Ripple has introduced GSmart, a set of AI agents built into its Ripple Treasury platform aimed at automating the routine work of corporate treasury management �� liquidity rebalancing across accounts, cash forecasting, risk monitoring and fraud protection, the company said.
The design draws a hard line between monitoring and execution. GSmart agents continuously analyze treasury data and can propose specific actions, but the platform requires a human to manually approve any action with material financial consequence before it executes — the agents surface recommendations rather than move money on their own. Ripple said governance rules for what the agents can and cannot propose are defined by a company's own treasury team through a companion tool called Knowledge Studio, where agents must cite the specific policy behind a recommendation before presenting it.
That human-in-the-loop requirement puts Ripple's treasury product on the more conservative end of a spectrum companies across the industry are now navigating: how much autonomy to grant an AI system that can see and recommend around a company's cash position, without letting it independently commit funds. Treasury management is a function where an erroneous automated transfer can be costly and difficult to reverse across banking rails, which is the underlying reason firms building in this space have generally kept a manual approval step rather than full autonomy.
Separately, Ripple also offers the XRPL AI Starter Kit, aimed at a different problem: enabling autonomous machine-to-machine payments using XRP and its RLUSD stablecoin directly on the XRP Ledger, without a human approving each individual transaction. Ripple positions that toolkit for lower-stakes, high-frequency machine payments, distinct from GSmart's higher-stakes, human-gated treasury actions — an indication that Ripple itself is treating "how autonomous should a financial AI agent be" as a question with different answers depending on what the money is for.
Ripple did not disclose specific customer adoption figures or cost savings tied to GSmart alongside the announcement, so its practical impact within treasury teams currently using it is not yet independently quantified.