The assessment format changelog

Graduate assessments changed faster in 2025–26 than in the previous decade — mostly because candidates started using AI, and employers started designing against it. This page is a dated, sourced log of what actually changed, and what each change means for how you prepare. Forum advice that predates an entry below is probably wrong.

Last verified 9 June 2026 · sources linked per entry
2025–26 cycle · PwC

PwC drops game-based assessment for an SHL suite

The Arctic Shores game-based "Career Unlocked" assessment is gone. PwC's 2025–26 process runs on SHL cognitive tests — numerical, verbal and inductive reasoning — plus a behavioural SJT and video interview.

What it means: PwC went from "unpreparable" games back to the three most trainable formats in existence. Timed numerical, verbal and inductive practice now directly moves your PwC outcome.
2024 → expanding through 2025–26 · BCG

BCG adds a live-proctored, calculator-free Cognitive Test

Built explicitly because candidates were pasting the Casey chatbot case into ChatGPT: 80 questions in 30 minutes (~22 seconds each), live camera proctoring over Zoom, pen and paper only. Debuted in Germany, now running in Austria, Switzerland, the Nordics and parts of APAC and Eastern Europe — with some offices running both Casey and the Cognitive Test.

What it means: raw mental arithmetic and logic at speed, with no tools, is back as a gating skill for consulting. The only preparation that transfers is volume under exam conditions.
2025 · Deloitte UK

Deloitte brings back face-to-face interviews to counter AI cheating

After the FRC flagged the integrity risk of fully-online graduate recruitment, Deloitte UK reinstated in-person interviews at the final stage. The online stages (Cappfinity immersive assessment + job simulation) remain.

What it means: an online score you can't reproduce in the room is now a liability across the Big 4. Train the skill, not the workaround.
2025–26 cycle · Linklaters

Linklaters re-administers the Watson Glaser in person

Candidates sit the Watson Glaser at home to progress — then sit it again in controlled conditions at the assessment centre. A material gap between the two scores speaks for itself.

What it means: the at-home test is an entry ticket, not the exam. Practise until your timed, unaided score is genuinely at the bar — that's the score that has to show up twice.
2025–26 cycle · Clifford Chance / KPMG / pupillage Bar

Explicit no-AI rules arrive, with teeth

Clifford Chance states AI use is not permitted on its Watson Glaser. KPMG requires a signed declaration that assessments were completed without AI — and warns that cheating can be reported to professional bodies. The Pupillage Gateway excludes AI-written applications from shortlisting via a sole-creation declaration.

What it means: "AI did my test" has moved from grey area to misconduct with a paper trail. The legitimate use of AI is on the training side — unlimited practice, instant feedback — which no declaration forbids.
Mid-2025 · McKinsey

Solve loses the Ecosystem game; Redrock and Sea Wolf carry the assessment

The famous reef-building Ecosystem game disappeared from many candidates' versions of Solve during 2025. The current core is the Redrock study (investigation → analysis → report, plus follow-up cases) and the Sea Wolf/Ocean Cleanup optimisation game, with a telemetry "process score" recording every click. An 85-minute variant adds a Sustainable Futures module.

What it means: most Solve walkthroughs on Reddit and YouTube describe a game you may never see. Check the date on any guide before trusting it.
2025–26 cycle · Bain

Bain's Sova test phases out verbal reasoning — and times you silently

The ~75-question Sova assessment (numerical, logical, personality) is dropping its verbal section for 2025–26. The test presents as untimed, but per-question response times are recorded and feed the score; some offices use TestGorilla instead.

What it means: "untimed" is a UX choice, not a scoring choice. Train at exam pace even when no clock is visible.
2022 → standard by 2025 · JPMorgan / HireVue

Pymetrics becomes HireVue games; facial scoring is long gone

HireVue acquired Pymetrics in 2022 — the 12 behavioural mini-games in JPMorgan's pipeline are now HireVue's game library. Separately, HireVue retired facial-expression analysis in 2021: current video-interview scoring is based on the language content of your transcribed answers.

What it means for video interviews: what you SAY is the score — structure, specificity, answering the actual question. Camera presence matters for humans who review later, not for the algorithm.
2025 cycle data · Civil Service Fast Stream

Fast Stream success rate hits ~1% on 72,691 applications

The 2024–25 cycle drew a record 72,691 applications for roughly 1,000 places. The current process: two untimed online tests (video/written work-based scenarios + a case study) inside a 7-day window, then a half-day virtual assessment centre — with feedback to every candidate at every stage.

What it means: at 1%, generic preparation is statistical noise. The differentiators are written analysis and judgement against the behaviours framework — both trainable, and the Fast Stream is unusual in telling you exactly where you fell short.
Survey data, 2025 · sector-wide

Employers double down on integrity as AI use doubles

The ISE's 2025 student recruitment survey: employers frequently encountering cheating doubled year-on-year to 15%; undisclosed AI use in interviews is the most-reported misconduct (61%); 79% are redesigning recruitment because of AI. Applications now average 140 per vacancy. One in ten students admits using AI in tests or interviews.

What it means: the arms race is real and the direction is set — more in-person verification, more proctoring, more formats that test the skill live. The candidates who win the next cycle are the ones whose ability survives the verification layer.

Train for the formats as they are now

Aren generates unlimited practice in the current formats — timed publisher-style tests, one-take video interviews, job simulations and a live group exercise — with instant feedback on every answer. One free timed test per format, no card.

Start practising free