- 90% of AI-using SMEs worry private data entered into their AI tool is being used to train it
- Over a quarter (28%) of SMEs leak company data into AI chatbots at least once a week
- 76% of SMEs would automate more if they had closed-loop AI that guaranteed data would never be used for model training
Startups.co.uk’s latest survey reveals that 90% of AI-using SMEs are worried that their company data is being used to train the AI tool.
This is a sensible apprehension, as company data entered into a public AI model can’t be recalled, deleted, or localized.
Company security traded for productivity
According to Startups.co.uk‘s findings, a staggering 50% of SMEs admitted to leaking company data into free public AI chatbots in order to complete work tasks.
Over a quarter of SMEs (28%) are leaking company data into AI chatbots at least once a week, and another 22% are doing so multiple times a month.
These businesses are effectively trading their long-term corporate security for short-term productivity.
The dangers of leaking company data into AI chatbots:
- Fines & operational disruption
IBM’s 2025 Cost of a Data Breach Report shows that a third (32%) of businesses who reported a data breach had to pay a fine and nearly all organisations suffered operational disruption.
Among organisations that reported recovery, most took more than 100 days on average to do so.
- Competitors could access your data
Proprietary workflows, pitch decks, and internal numbers may already be sitting in a competitor’s future AI prompt response.
Despite widespread concern for company security, startups continue to upload data in order to keep up with daily workloads.
Manual AI security checks waste company time
Startups.co.uk found that businesses are manually preparing data before using AI tools, to help mitigate security risks.
Almost a quarter (23%) of businesses say they spend a ‘significant’ amount of time manually preparing data, meaning it often takes longer to clean the data by hand than AI takes to process it.
Additionally, over half (52%) of leaders who use AI spend a moderate amount of time manually preparing or anonymising data to ensure privacy and security.
This need for manual preparation and rework is a clear downside of public AI adoption. Those who do not use public AI face far less friction: only 11% report significant manual rework compared to 22% of public AI users.
Editor of Startups.co.uk, Zohra Huda, comments:
“Our data shows how startups are caught in a tough productivity trap. Half of all SMEs leak data into public chatbots just to keep up with daily workloads, yet 90% are privately terrified of where that data will end up. UK founders know that blindly feeding proprietary data into a public model means their pitch decks or internal numbers could easily end up in a competitor’s future AI prompt but they feel they don’t have a choice. Is the price of innovation gambling with your company’s intellectual property?”
