🔍 Read the full analysis: How To Understand UBS’s Personal AI Agent Experiment on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
UBS reported more than 12,700 AI personal agents run by employees in its third-quarter results on October 28, 2026. The bank has not disclosed what those agents can access or do, or how they are monitored; its CEO separately said outside AI models are not allowed to act for clients.
UBS said employees were running more than 12,700 AI personal agents to support their daily work, putting a concrete figure on the Swiss bank’s use of workplace AI. The October 28 disclosure also listed 608 AI applications in production and more than 1,090 in development, but did not explain what the agents can access, what actions they can take or how staff activity is reviewed, a concern illustrated by an AI agent experiment that revealed hidden files.
The figures appeared in UBS’s third-quarter results release, under a section on investment in the bank’s business. UBS reported nine large-scale AI initiatives, 608 live AI use cases—an 81% increase from a year earlier—and more than 1,090 additional applications in development. It also said over 18,000 software developers were using AI tools in their work cycles, describing the productivity gains as measurable without quantifying them.
UBS reported that its STAAT tool had generated more than 30 million insights so far in 2026 for the bank’s US financial-adviser teams. Reuters reported the agent figure in its October 28 coverage of the results, and PYMNTS reported it the following day. Neither report supplied further detail about the agents’ design or use.
The agent count is not a count of employees, and UBS did not say how many staff members use them or whether each employee has one. The bank’s July 29 second-quarter release said all employees had the ability to create their own AI agents, part of what may become a new personal agent layer. That describes access to a capability, not the number of people who have created or actively use an agent.
Infographic 01 · What UBS reported
12,700 personal AI agents, inside one bank
The AI figures from UBS’s third-quarter results, and how fast they grew in three months.
From permission to 12,700 agents
UBS 2Q26 release (29 Jul 2026) vs. 3Q26 release (28 Oct 2026)
What the agent number is not
- Not 12,700 people. One employee can run several agents; UBS does not say how many people use them.
- Not a definition. UBS does not say what an agent can access or do on its own.
- Not a result. “Measurable productivity gains”, but no hours or costs published.
- Scale, for context: about 13 agents per 100 internal full-time staff (12,700 ÷ 97,593, own calculation).
“Our employees are running over 12,700 AI personal agents to support them in their daily activities.”UBS, third-quarter 2026 results release, 28 October 2026
Also in the release: 9 large-scale AI initiatives and more than 30 million insights from the STAAT tool for US financial advisor teams so far this year. UBS placed 6th of 50 banks in the 2026 Evident AI Index.
Sources: UBS 3Q26 release, 28 Oct 2026 · UBS 2Q26 release, 29 Jul 2026. Reported identically by Reuters, 28 Oct 2026 and PYMNTS (29 Oct 2026).
Powered by Thorsten Meyer AI
Why Agent Oversight Matters at UBS
The disclosure gives businesses a measure of how quickly employee-facing AI can spread inside a large organization. It also puts a practical governance issue in view: the count alone says little about the level of risk. A tool that summarizes a document for an employee is different from one that can send a message, update a record, approve a request or move money.
That distinction matters especially at a bank, where staff handle sensitive information and decisions can affect clients and financial records. UBS has not said whether its personal agents can take actions independently or only prepare work for a person to check. Without details about permissions, data access, human review and incident controls, readers cannot tell from the headline number how autonomous or consequential these systems are.
The disclosure is also separate from any claim that AI has reduced UBS staffing. The bank’s internal workforce fell by about 1,500 full-time equivalents during the quarter, to 97,593; UBS and Reuters attributed the decline to the Credit Suisse integration. UBS did not connect the headcount change to AI, and the available reporting does not establish such a link.
Popular right now
TCL QM64L 85-Inch QD-Mini LED 4K Smart Fire TV
Shark Rotator Lift-Away Upright Vacuum with HEPA Filter (NV501)
Ninja DualZone Smart XL Air Fryer, 10-Qt, 6-in-1, DZ950
As an affiliate, we earn on qualifying purchases.
From Staff Access to Agent Counts
In its July 29 results, UBS said employees could create their own agents, describing the program as a way to expand AI use while allowing staff to spend more time on judgment, collaboration and client work. The October update supplied a usage figure but did not explain how the bank defines a “personal agent.” The term remains undefined in the results release.
UBS’s technology materials describe other employee AI tools, including an in-house assistant called Red, made available to around 100,000 employees, and an HR assistant in Microsoft Teams for tasks such as booking leave and finding policy information. Those examples offer some context for workplace AI at the bank, but UBS has not said they are among the 12,700 agents or that all personal agents work in the same way.
There is a separate boundary around AI acting directly for customers. On October 27, CEO Sergio Ermotti said UBS was not allowing outside AI models to enter the bank and act for clients, such as managing accounts or moving deposits, according to Reuters. That statement concerns external models acting on clients’ behalf; it does not, by itself, define the permissions available to UBS employees’ internal agents.
“Employees are running over 12,700 AI personal agents to support them in their daily activities.”
— UBS, in its October 28 third-quarter results release
What UBS Has Not Disclosed
The results release does not define a personal agent or state which models and tools power them. It does not detail what internal systems or data the agents can access, whether they can perform actions without employee approval, or what review and monitoring processes apply. UBS also did not report how many employees use agents, how often they are used, or whether any are client-facing.
Those omissions limit what can be concluded from the figure. The reported count confirms that UBS says a large number of employee agents are running; it does not establish that all are autonomous, that they make decisions, or that they have a direct effect on client outcomes. Nor does the release quantify the productivity gains associated with developers’ use of AI tools.
Swiss financial regulator FINMA expects banks to maintain an inventory of AI applications and classify them by risk, according to the source material. The regulator has said keeping that inventory has become harder as applications become widely accessible. The available account does not specify how UBS’s 12,700 agents are entered into that inventory or what risk classifications the bank applies.
The Next Test Is Governance
UBS’s results provide a snapshot, not a full account of how the agents operate. Further detail from the bank on permissions, human approval, data access and monitoring would help staff, clients and investors understand what the reported systems actually do and how the bank manages related risks. The company has not, in the material available, announced a date for a fuller explanation.
For now, the key distinction to watch is between agents that prepare information for employees and agents that can carry out consequential actions. UBS’s statement about excluding outside models from acting for clients addresses one boundary, but it does not answer questions about the capabilities of internal employee tools. Any future figures should also be read with their definitions and time periods: a rising count alone would not reveal how many agents are active or how much work they perform.
Key Questions
What did UBS report about its AI agents?
UBS said that employees were running more than 12,700 AI personal agents to support their daily activities in its October 28, 2026, third-quarter results release.
What can the UBS agents do?
The results release does not define their capabilities. UBS has not specified their data access, whether they can take actions independently, or how much employee review is required.
Are the 12,700 agents 12,700 employees?
No such link was reported. UBS gave a count of agents, not users, and did not say how many employees created or actively use them.
Did UBS say AI caused its workforce to shrink?
No. UBS’s workforce fell by about 1,500 full-time equivalents in the quarter, to 97,593. UBS and Reuters attributed the decline to the Credit Suisse integration; UBS did not link it to AI.
Can outside AI models act for UBS clients?
CEO Sergio Ermotti said UBS was not allowing outside models to enter the bank and act on clients’ behalf, including managing accounts or moving deposits, according to Reuters. That statement does not explain the permissions of UBS’s internal employee agents.
Source: ThorstenMeyerAI.com
