BlackFog has published research showing that 47% of security leaders have experienced sensitive data exposure through AI tools. The survey covered IT security decision-makers at large companies in the UK and US.
The findings suggest governance and technical controls are struggling to keep pace with AI adoption in businesses. Among organisations that had experienced a data incident, 77% said a sanctioned, company-approved AI tool was involved.
BlackFog commissioned a survey of 400 IT security decision-makers at companies with more than 500 employees, split evenly between the UK and the US. It found that 39% of respondents said AI adoption was moving faster than their organisation could govern, while only 27% of those who felt outpaced had comprehensive AI controls in place.
Almost all respondents, 98%, said their organisation had an AI usage policy. Enforcement was less consistent: 35% said they had a written policy setting out what employees must not enter into AI prompts but no technical enforcement, while 56% said they supported broader AI usage policies with comprehensive technical controls.
Visibility gaps
The research also highlighted limited oversight of the data employees share with AI systems. While 62% of respondents said they could identify the users or departments sending data to AI tools, only 48% said they could provide any record of the types of data that had left the organisation through those tools.
The survey pointed to a similar gap around autonomous AI agents. More than half of respondents, 56%, said AI agents acting on behalf of employees fell outside their organisation's existing controls. In the technology and telecoms sector, that figure rose to 70%.
It also identified shadow AI as an area where detection is stronger than prevention. Only 54% of respondents said they could block or restrict unsanctioned AI tools, while 32% said they could detect shadow AI use but could not consistently stop it.
Weakest controls
The weakest controls were around proprietary design, creative and marketing content. In that category, 32% of respondents said they could detect or monitor sharing but could not consistently prevent it, while a further 20% said they relied mainly on written policies or training.
Across other forms of potentially sensitive material, including sales data, proprietary code and intellectual property, only about half of organisations said they could technically prevent risky sharing behaviour.
Dr Darren Williams, Founder and Chief Executive Officer of BlackFog, said the results suggested approved AI services could pose as much concern for security teams as unauthorised ones when controls over data inputs are weak.
“Among organisations that reported sensitive data exposure, 77% said a company-approved AI tool was involved. This shows why security teams cannot stop at deciding which AI tools employees are allowed to use. Organisations also need to control what information is shared with them.
“Written policies set expectations, but they cannot prevent sensitive data from leaving the organisation on their own. Businesses need visibility into the data being shared and technical controls that can stop unauthorised disclosure at the point it happens. Those protections need to cover approved AI tools, unsanctioned services and autonomous agents acting on employees' behalf.”