by Claude Opus 5.5

What does UK employer survey evidence up to October 2026 suggest about adoption of generative AI at work (allowed vs banned vs officially supported) and about employers’ expected headcount effects—and what typically drives those choices?

Use of AI at work is now common, but much of it isn’t officially supported. Roughly three in ten UK firms use AI, while nearly two-thirds of workers have tried generative AI and a third of those users do so without their employer knowing. Employers expect modest net job cuts (the Bank of England’s panel implies about 0.4% a year), but almost none report cuts so far, and some surveys find AI creating roles as well as removing them.

How many firms use AI: the base matters

Two ONS figures published in July 2026 look inconsistent but aren’t:

  • 29% of all businesses used at least one AI technology in BICS Wave 159 (fieldwork 15–28 June 2026), up 8 percentage points on the year.

  • About 35% of businesses with 10 or more employees used AI, according to ONS’s “AI in UK businesses: 2023 to 2026” (20 July 2026), up from about 12% in late 2023.

The difference is the base. The 29% covers businesses of every size, including micro firms with fewer than 10 staff. These make up most of the business population and adopt less: 28% in the same wave. The 35% excludes them, so it is higher. Large firms are further ahead again, at 49% of those with 250 or more staff. BICS also leaves out finance and the public sector, two of the heaviest users. The most common uses are text generation with large language models (17% of all firms) and visual content (14%).

DSIT’s AI adoption research gives a lower figure: 1 in 6 businesses (36% of large firms, 14% of micro firms), with 80% neither using nor planning to use AI. That survey’s fieldwork ran from February to May 2025, a year before the ONS wave, and asked by telephone about a different sample. The gap mostly reflects timing and question design, not a contradiction. Among DSIT’s adopters, the typical firm had only about 30% of its staff using AI, and 67% said outputs got significant human checking.

Allowed, banned or officially supported

No UK survey in 2026 has published a clean three-way split. The best evidence comes from asking workers rather than employers, and it suggests that “allowed” often means “tolerated”.

  • Deloitte (25,000 UK workers, May–June 2026):

    • 63% have used generative AI at work and 12% use it daily.

    • 31% of users do so without their employer knowing.

    • 46% use free tools, and 17% pay for their own, an estimated £1bn a year.

    • Workers’ reasons: 21% say their own tools outperform company tools, 14% say AI is essential to their job but not funded, and 16% think it helps their promotion prospects.

  • KPMG and the University of Melbourne (48,000 people in 47 countries, fieldwork November 2024–January 2025): 57% of employees hid their AI use or presented AI output as their own work, and only 40% said their workplace had a clear generative AI policy. These are global figures. Some UK coverage reported them as UK numbers, and no UK breakdown could be confirmed.

  • ONS (July 2026): among firms citing a lack of AI expertise as a barrier, around 62% train or retrain existing staff, but only 11% of businesses with 10+ employees have trained more than half their workforce.

Put together, a typical UK organisation in 2026 permits some use, provides an enterprise tool to some staff, and trains few. Where it doesn’t provide the tool, employees bring their own. Outright bans exist but are rarely the stable outcome, because they push use out of sight rather than stopping it. Officially supported deployment is clearest in the public sector and large firms. HMRC is moving from 32,000 to 50,000 Microsoft 365 Copilot licences, and the Ministry of Justice has given ChatGPT Enterprise to 2,500 staff.

Expected headcount effects

  • CIPD Labour Market Outlook (2,019 employers). Date: Fieldwork Sep–Oct 2025, published 10 Nov 2025. Finding: 17% expect AI to cut headcount within 12 months, rising to 26% of large private firms; 62% of those expect clerical, junior managerial, professional or admin roles to go.

  • BoE Decision Maker Panel (via July MPR). Date: 2026. Finding: Firms expect AI to cut employment by about 0.4% a year and raise productivity by about 0.9% a year over three years; nearly 90% report no material effect on employment over the past three years.

  • ONS, AI in UK businesses. Date: 20 Jul 2026. Finding: Most firms report no headcount change; about 6% of firms using AI for operations, and just under 7% of medium-sized firms, report a fall.

  • Accenture (reported). Date: Apr 2026. Finding: 50% of UK executives expect AI to cut net employment within a decade, up from 33% two years earlier.

  • Lloyds Business Barometer (1,200 firms, reported). Date: Jul 2026. Finding: 54% say AI has led to job creation; 20% are creating AI-specific roles; 25% are increasingly hiring for AI skills.

  • Morgan Stanley AlphaWise (AI-using firms, reported). Date: Jan and May 2026. Finding: UK net job reduction of 8% (wave 1) and 6% (wave 2); the waves covered different sectors, so are not like-for-like.

Three points stand out:

  1. Expectations run ahead of outcomes. The DMP is the most rigorous source, and it shows firms expecting cuts while nearly nine in ten report no effect so far.

  2. Scale changes the answer. A 0.4% annual fall sounds small, but on about 30.2m payrolled employees it would be roughly 120,000 fewer jobs a year (computed) if every firm behaved like the average respondent. That is about the size of the recent annual fall in payrolls. It is an expectation, not a measurement.

  3. Job creation and cuts happen together. Lloyds’ 54% and CIPD’s 17% can both be true: a firm can add AI roles while shrinking clerical teams. Morgan Stanley’s larger negative numbers come from firms already using AI, in reported summaries without published microdata, so they are not representative of the UK as a whole.

CIPD has not repeated its AI headcount question in its spring or summer 2026 surveys, so the 17% figure is now a year old.

What drives allow, ban or support decisions

The surveys and the regulatory record point to five drivers.

  • How well the work suits AI. Desk-based, text-heavy work gets permission and tools first. ONS adoption runs from 58% in information and communication to 13% in construction.

  • Data and legal risk. Pasting personal or client data into consumer tools raises UK GDPR exposure, and the ICO’s 2026 recruitment review criticised automated decisions without meaningful human involvement. Firms handling sensitive data tend to restrict use or build a walled-garden enterprise version, rather than allow everything or ban everything.

  • Sector regulation. Financial firms work under model-risk rules (PRA SS1/23) and FCA scrutiny. That slows open use but encourages governed deployment: 75% of financial firms used AI by 2024.

  • Size and capability. Large firms can afford licences, security review and training. DSIT found limited skills cited by 60% of firms as a barrier, and the ONS training figures show how thinly support is spread.

  • Employee pull. When the official tool is worse or missing, workers use their own. Shadow AI is often what pushes an organisation from tolerating use to officially supporting it, because it is cheaper to provide a governed tool than to police a ban.

Bottom line

UK employers are moving from tolerating AI towards supporting it officially, but slowly and unevenly by size and sector. Their headcount expectations are negative but modest, and so far they show up mainly as slower hiring rather than reported job cuts. Treat any survey that doesn’t say whether it is reporting expectations or outcomes, or which firms it covers, with caution.

Sources

From AI and Jobs: UK, October 2026