UBS puts AI proficiency into the graduate interview for its 2027 intake

A recruiter with forty minutes and a scoring sheet has to turn "use and experiment with AI responsibly to improve business outcomes" ^[3] into a question that has a better and a worse answer. Nothing in the reporting describes the rubric.

What candidates actually bring to that question is a story about summarising reading for a dissertation, or a spreadsheet macro a model wrote for them that worked. What recruiting teams tell themselves they are buying is judgment: someone who knows which tasks the model is bad at and checks the output before it reaches anyone else. Those two candidates sound the same at the start of an interview, and the confident one usually sounds better.

The grid worth drawing has tool fluency on one axis and verification habit on the other. High fluency with no verification habit is the quadrant that interviews best and costs the most later, because that hire moves fast and produces work nobody downstream can retrace. Low fluency with a strong verification habit is trainable inside one internship. The question that separates them asks about a task the candidate decided not to use a model for, and how they decided. That is answerable by someone with real practice and almost unfakeable by someone with a ChatGPT tab and good patter.

TechRadar sets the hiring change against a Morgan Stanley report from early 2026 warning that 200,000 banking jobs could go in Europe over five years ^[8], which averages 40,000 a year ^[11]. The same piece reads the graduate requirement as evidence that AI is not replacing entry-level workers, with UBS treating it as a productivity aid for human staff ^[9], and suggests junior responsibilities drift toward managing AI output ^[10]. No UBS intake figure appears anywhere in the material ^[13], so the Europe-wide number and the bank's own hiring cannot be set against each other. Both readings survive if the reductions land somewhere other than the graduate programme.

As for the idea that regulated employers will copy this, the material names no other one doing it ^[12]. What is portable is not the requirement but the honesty underneath it: TechRadar reads UBS as asking for a level of proficiency and a willingness to learn rather than the finished article ^[7]. An employer that teaches its own domain cases after the offer is admitting the screen can only test disposition. That is a reasonable thing to test, provided the interview question is built for it rather than borrowed from the press release.