UBS has added a new requirement to its junior hiring: from the 2027 intake, applicants to Global Banking and Markets will be expected to demonstrate not just that they have used AI, but that AI improved something they produced. According to reporting by the Financial Times and Switzerland's NZZ am Sonntag, the bank wants candidates to show a real task where a model beat their prior output, with the improvement measured in outcomes and efficiency. Touching ChatGPT is not enough.
The distinction matters because it changes what the interview is testing. A fluency check asks whether a candidate knows the tools. This bar asks for evidence of a result. The applicant is not claiming competence with software. They are claiming the software made their work better, and they are expected to prove it.
The test is output, not fluency
The reported policy applies to graduate and intern hiring into Global Banking and Markets, the division that houses investment banking, sales and trading, and research. The FT, citing people familiar with the matter, reported that the requirement will sit alongside existing academic criteria rather than replace them, and that AI literacy "complements, rather than replaces" the analytical and interpersonal skills the bank has always screened for. UBS told the FT that "AI capabilities and experiences have become an important aspect of future professional success."
Interviews for graduate trainees and interns will include questions on AI fluency, new job advertisements carry the requirement, and successful hires enter an internal "AI Fluency Pathway" covering banking use cases and responsible use. The NZZ attributed the initiative internally to group human resources chief Stefan Seiler, a member of the executive board, who declined to confirm the newspaper's account. Reuters had not independently verified the plan as of September 7.
What is unusual is the standard itself. Most skills gates ask whether a candidate can do something. This one asks whether a machine improved what they did, which forces applicants to think in terms of before and after: their baseline work, the model's output, and the measurable difference between them. A candidate who cannot articulate that comparison fails the test no matter how many AI tools they can name.
The entry bar rises as the door narrows
The timing gives the policy its edge. UBS employed 99,085 full-time staff at the end of June 2026, below the 100,000 mark and down from 103,200 at the end of 2025. Roughly 3,000 further redundancies tied to the Credit Suisse integration are still expected, concentrated in the second half of 2026 and early 2027, while chief executive Sergio Ermotti has said the integration should be substantially complete by the end of 2026.
So the bank is tightening headcount and raising the entry credential at the same time. Juniors will compete for fewer seats under a higher bar, and the bar itself is about the technology that is widely forecast to shrink the junior analyst's role. Morgan Stanley research has estimated that 200,000 European banking jobs are at risk over five years, and McKinsey has said junior analyst intakes could fall by as much as two-thirds as AI takes over the work those analysts once did. UBS is asking the class of 2027 to prove their AI judgment at the exact moment the industry is deciding how many of them it needs.
That overlap is not necessarily a contradiction. A bank can believe AI changes the analyst job without believing it eliminates the analyst, and UBS has said the requirement complements traditional skills rather than replacing them. But the two movements pull in opposite directions, and graduates applying this cycle will feel the squeeze between them.
The bank is hiring to where it already is
The hiring bar mirrors an internal buildout that has been underway for more than a year. UBS named its first chief AI officer, Daniele Magazzeni, in January 2026, and has reported more than 300 AI use cases deployed during 2025, a firm-wide Microsoft Copilot rollout, and an internal large language model the bank calls Red. Around 35 of its 720 research analysts now have AI avatars that present their work. Through July 2026, the FT reported, roughly 38,000 employees had completed AI learning journeys, 49,000 were enrolled in AI courses, and 21,000 had earned internal "AI Citizen" badges.
A hiring requirement that candidates arrive with demonstrated AI ability is the logical endpoint of that investment. If the bank's workflows assume staff can direct a model toward a better outcome, then a graduate who has never produced one is arriving behind the organization rather than joining it. The new bar does not ask applicants to be unusual. It asks them to be where the bank already is.
The industry context is moving the same way. JPMorgan made prompt engineering training mandatory for new hires in 2024, and its chief executive, Jamie Dimon, said this year that "we will hire more people in artificial intelligence and fewer bankers in certain categories." Santander now seeks "advanced AI users" in parts of its graduate program. UBS is not alone in the direction. It is alone, so far, in asking for a concrete before-and-after improvement as the evidence.
The case against the new bar
Skeptics argue the standard measures the wrong thing. Conor Hillery, JPMorgan's co-head of EMEA investment banking, warned in December 2025 that "juniors cannot afford to lose the fundamentals," and academic research on what some call "never-skilling" suggests that screening for AI proficiency does nothing to protect fundamentals and may even select against them, by favoring candidates who lean on the model rather than the skills beneath it. Goldman Sachs has taken the opposite posture in its own hiring process, banning ChatGPT use during interviews so it can see how candidates perform unaided.
There is also a simpler reading, aired in the Swiss press: that an AI requirement is a convenient filter in a cycle of layoffs, dressing headcount pressure in the language of technological progress. One academic quoted in coverage of the policy, NYU Stern's Robert Seamans, has described AI as a "narrative tool" that can mask macro-driven cuts. UBS would not confirm the initiative through official channels, which leaves room for both interpretations.
The bank's answer, implicit in its statement, is that the requirement is a complement, not a replacement. Academic results, interpersonal skills, and analytical ability still matter, and a candidate who shows a model beating their prior work still has to explain the task, the method, and the difference. That explanation, not the model's output, is the part a competent interviewer grades.
The credential has changed
Hiring tests used to verify what a candidate knows. UBS is testing for something adjacent: whether a machine made their work better, and whether they can prove it. The reference point is no longer the other applicants. It is the candidate's own prior output.
That is a different kind of screening, and its long-term effect depends on which candidates it rewards. If it selects for people who understand when a model helps and can measure the difference, banks get analysts who treat AI as a tool with a margin of error. If it selects for people who lean on the model and perform the explanation, banks get the "never-skilled" juniors the skeptics warn about. The first cohort measured against the new bar will not answer that question. Their careers will.
Primary sources
- Financial Times reporting, syndicated by CNA for the 2027 intake policy and the training figures.
- The Next Web for the policy details, industry context, and the skeptics' case including Conor Hillery's warning.
- FStech for UBS's statement and the analyst avatar figures.
- Banking Dive for UBS's AI buildout under chief AI officer Daniele Magazzeni.
- Swissinfo for UBS headcount and the Credit Suisse integration timeline.