White-collar work used to be thought of as relatively safe from automation. Computers could process transactions, but people still had to research the issue, draft the document and explain the answer. Generative AI is changing that division of labour.
It is tempting to rank entire professions from "safe" to "doomed". That would be misleading. The relevant unit is usually the task, not the job title.
What does "at risk from AI" actually mean?
There are at least three different questions: Can an AI system attempt the task? Can it perform the task reliably with the organisation's data and constraints? And would using it actually reduce demand for employees?
Those questions often produce different answers. Anthropic's March 2026 research distinguished theoretical AI capability from observed use, and reported no systematic rise in unemployment in highly exposed occupations in its study, while flagging suggestive signs of slower hiring for younger workers.
High exposure is not a confirmed redundancy forecast.
The white-collar tasks attracting the most attention
Administrative processing
Data entry, routine correspondence, categorising requests and transferring information between systems are obvious candidates. The work often follows a pattern. The challenge is what happens when a case does not fit the pattern.
Research and first drafts
A model can assemble an initial landscape review, summarise lengthy material or prepare a draft email. It is less dependable at deciding whether the sources are trustworthy, what has been omitted or which conclusion the organisation should adopt.
Financial and operational reporting
Recurring commentary, variance descriptions and straightforward spreadsheet analysis can be accelerated. Incorrect assumptions, broken formulas and unauthorised data access remain very real risks. Someone must own the numbers.
Standard legal and compliance documents
AI can help extract clauses, compare versions and create initial summaries. That does not make it responsible for legal advice, privilege, final sign-off or the facts of a particular client matter.
Entry-level analytical work
Graduates and junior analysts often learn by doing the same first-pass tasks AI can assist with. Firms may change the way they train and staff these roles. Losing that learning pathway could become a problem even where senior roles remain.
A quick comparison across professions
Accounting: Routine reconciliations, document handling and commentary may become quicker. Controls, tax positions, audit judgement and client advice still need accountable professionals.
Law: Document review and first drafts are exposed. Negotiation, advocacy, context and professional obligations place limits on delegation.
Banking: Research packs, presentations and analysis may move faster. Transaction judgement, risk ownership and relationships are much harder to automate end to end.
Consulting: Information gathering and slide production are changing. Deciding what the client should actually do, securing agreement and delivering results remain central.
B2B sales: Prospect research and outreach preparation are changing. Trust, discovery, complex negotiation and account strategy still matter.
These are patterns of task exposure, not promises about job security. We cover each profession in more detail in this series.
The risk that gets overlooked: fewer new vacancies
Automation does not have to lead to a mass layoff to change a career. If a team decides that three people can do work once performed by four, the next vacancy may simply stay empty.
That is particularly important for people trying to get their first professional job. They need opportunities to acquire judgement, not just a list of software tools to learn.
How to work out where you stand
Take a recurring deliverable and divide it into five stages: gather information, structure it, produce a draft, verify it and secure a decision. Mark where AI could assist safely, where review is essential and where the work depends on relationships or accountability.
Then ask what capability you could develop if you spent less time on stage one. Better problem definition? More client contact? Stronger analytical checking? That is where a sensible career response begins.
Read how AI could affect graduate jobs and our practical 30-day plan next.
