UBS bars external artificial intelligence models from taking actions for clients, Chief Executive Sergio Ermotti said Tuesday, citing risks that systems could manage accounts or shift deposits.
Reuters reported that Ermotti spoke at a science conference but did not identify the event. He said UBS was not allowing models to “enter into the bank and do that,” referring to actions on behalf of clients.
UBS describes AI applications run within the bank. It says technology partnerships broaden the models and tools available to its teams. The bank’s responsible-AI policy says it aims to preserve people’s ability to make decisions, limit harm and apply fair, transparent processes.
UBS says its CORISC XAI tool screens transactions and automatically handles about 20% of alerts under internal oversight. The bank’s U.S. equity order-handling procedures, current as of July 2026, say AI supports trading algorithms and its Smart Order Router, also under internal oversight. They also permit UBS to use client orders and related data when training or retraining its own models.
UBS’s second-quarter results, released July 29, counted 560 AI use cases in production, up 92% from a year earlier, more than 920 applications in development and nine large-scale AI initiatives. The bank’s public materials describe these internal uses but do not explain how Ermotti’s restriction applies to its model partnerships or internal processes.
Reuters reported Oct. 22 that major U.S. banks had moved to restrict third-party agents’ ability to transact through customer accounts, citing two people familiar with the matter. A person in the banking industry told Reuters that transactions made using credentials customers shared with agents could not be deemed unauthorized.
Reuters also reported that Ermotti linked the pace of UBS’s AI rollout to work integrating Credit Suisse. EconoTimes previously covered the integration’s effect on UBS technology work.
