Ripple bets $1B on AI-powered treasury

Ripple puts AI agents inside its $1 billion corporate treasury bet

Ripple is expanding its corporate treasury software with new AI “agents” designed to help finance teams monitor liquidity, forecasts and financial risk, and then recommend what to do next. The company says humans will still have to approve every action before any transaction is executed.

The new tools, branded GSmart, are being added across Ripple Treasury, the treasury-management platform Ripple gained through its roughly $1 billion acquisition of GTreasury in October last year. Ripple described the rollout in a Thursday release.

According to the company, the AI agents can track a firm’s cash position and risk exposure, surface relevant policies, and suggest specific steps a treasury team could take. Ripple also emphasized guardrails aimed at reducing errors: the agents rely on a company’s own rules for recommendations, and deterministic software continues to handle financial calculations, rather than leaving core math to generative AI.

The update underscores Ripple’s broader push to expand beyond its long-standing focus on crypto payments and digital asset operations. By buying GTreasury and rebranding it as Ripple Treasury, Ripple moved into a market typically served by specialist providers that manage cash positioning, forecasting, reconciliation and reporting for corporate finance departments.

AI features have become a growing theme in enterprise finance software, where companies want faster monitoring and decision support without giving up control over approvals and compliance. Ripple’s approach follows that pattern by using AI for analysis and recommendations while keeping execution under human oversight.

• What changed: Ripple added GSmart AI agents to Ripple Treasury to monitor cash, liquidity, forecasts and risk and recommend actions.
• Why it matters: It strengthens Ripple’s post-acquisition treasury strategy and positions its platform as more automated without removing human control.
• Key constraint: All actions still require human approval, with deterministic software handling calculations to limit AI-related errors.