One in six British workers is paying for an artificial-intelligence tool to help with work, collectively spending nearly £1 billion a year from their own pockets, according to a Deloitte survey.
The inaugural GenAI Workforce Survey, based on responses from 25,000 workers gathered by Ipsos, found that two-thirds had tried tools such as ChatGPT, Claude, Gemini or Copilot. Nearly a quarter used generative AI daily at work, while 17% paid for at least one tool themselves.
The result describes an unusually enthusiastic workforce and an unusually absent expenses department. Employees have apparently responded to the AI revolution by becoming their own procurement team, security reviewer and monthly direct debit.
The rise of shadow AI
Deloitte found that 31% of workers used generative AI without their employer’s knowledge. This is often called “shadow AI”: tools adopted outside approved systems, policies and oversight.
The attraction is obvious. Respondents said AI saved an average of 70 minutes a week, most commonly by searching for information, drafting emails and creating summaries. For an employee confronting a blank document and twelve unread meeting notes, a £20 subscription can look less like a luxury and more like emergency office equipment.
The risk appears when confidential material travels with the prompt. Staff may paste client data, contracts, source code, financial figures or internal discussions into a consumer account without knowing how that information is retained or used. Even when a provider has strong controls, the employer may have contractual and legal obligations that a personal subscription cannot satisfy.
Accuracy is another problem. A polished answer can still contain invented facts, missing context or insecure code. Saving 70 minutes is less impressive if somebody later spends Friday afternoon explaining why the confident summary described a meeting that never happened.
Employers are behind the behaviour
The survey suggests workers are adopting AI faster than organisations are providing approved tools. Simply banning it is unlikely to work when employees can reach powerful services from a phone. Better governance means approved accounts, clear rules about data, training, human review and a way for staff to request useful capabilities without completing a procurement pilgrimage lasting three financial years.
There is also a fairness question. Workers who can afford premium tools may gain an advantage over colleagues doing the same job. If software is genuinely necessary for productivity, the employer should usually provide it, just as it provides laptops and licensed spreadsheets. “Bring your own intelligence subscription” is not a mature workforce strategy.
Deloitte says organisations will gain most when they move beyond access and help people use AI with purpose and guardrails. That is less exciting than announcing an innovation lab, but it is where benefits and risks become measurable.
For the wider policy argument, see our coverage of the UK royal AI safety summit.
What workers should do now
Check the employer’s acceptable-use policy before uploading work material. Avoid personal or confidential data, verify important outputs, keep records of human decisions and ask whether an approved enterprise account exists. If a manager expects AI-assisted productivity, ask whether the subscription can be expensed and the training supplied.
The OutOut verdict
British workers did not wait for the corporate transformation webinar. They entered card details and carried on with Wednesday.
That initiative is useful evidence that AI has become ordinary workplace infrastructure. It is also a warning. Companies cannot demand faster output, ignore the tools producing it and then look astonished when sensitive documents visit an account registered to somebody’s Gmail address. If the robot is now on the team, HR and IT should at least know its name.
