The billion-pound vacuum: Why shadow AI is forcing businesses to act
One in six employees pay for their own GenAI tools and some bring them to work
Digital transformation is experiencing a crisis of control. According to a recent study by Deloitte, UK employees are expected to spend an estimated £1 billion of their own money on unregulated, privately funded AI tools.
The phenomenon of “shadow AI” exposes a deep divide between employees’ appetite for innovation and the speed at which corporate leadership is responding. For CIOs and CISOs, this is no longer simply a compliance problem, but a serious operational and strategic security risk.
According to a Deloitte survey, generative AI is already finding its way into everyday working practices outside official IT structures. Almost one in three users, 31%, uses these applications without their employer’s knowledge. One in six, 17%, pays for the tools personally, with this private expenditure estimated to total £1 billion a year in the UK alone. Meanwhile, 46% of workplace GenAI users rely on free services.
The figures demonstrate both the scale of demand for productive AI tools and the urgency with which organisations need to provide transparency, secure alternatives, and clear rules governing data protection and control.
Staff are under continuous pressure to perform and save time. If the official IT environment does not provide adequate AI-enabled tools, employees become more likely to turn to freely available consumer applications.
This pursuit of efficiency exposes an administrative vacuum at leadership level. Many executives are reluctant to establish clear budgets and frameworks for AI use, with potentially serious consequences.
The accessibility of modern AI services, which can be used through simple web interfaces on almost any platform, effectively bypasses traditional software approval processes and endpoint controls.
The changing attack surface
For security leaders, shadow AI changes the attack surface of the entire organisation. The primary risk is the uncontrolled loss of intellectual property, internal source code, financial data or personal information.
Once such information is entered into freely available versions of large language models, it may be processed and retained outside the organisation’s direct control.
Shadow AI also operates largely through the browser. Conventional firewalls and security architectures are often not designed to analyse the semantic content of prompts. They may recognise legitimate HTTPS traffic to a known web address, but not the sensitive information being transmitted.
Security incidents involving shadow IT or unregulated AI produce significantly higher follow-on costs than breaches of conventional IT infrastructure. According to an IBM report, the average cost of a data breach at organisations with high levels of shadow AI was $670,000 greater than at organisations with little or no shadow AI.
Strategic lessons for CIOs and CISOs
Deloitte’s figures highlight the limitations of prohibition-only strategies. Where strong demand encounters regulatory barriers, an unofficial market for IT use inevitably emerges.
For CIOs and CISOs, this creates four key areas for action if they are to regain control of corporate data flows.
Move from gatekeeper to enabler by providing secure alternatives
The most effective way to curb shadow AI is to offer an attractive, controlled alternative. CIOs should proactively provide enterprise licences for approved AI platforms.
Enterprise versions can include contractual commitments governing where submitted data is held and whether it is used to train publicly available models.
Establish dynamic AI governance policies
Rigid, multi-year IT policies are ineffective for a technology that develops in cycles measured in weeks. Organisations need lean and agile governance structures.
CISOs should draw up clear policies that focus not primarily on prohibition, but on defining the boundaries of safe use. Employees are looking for this guidance: surveys regularly show that a broad majority of workers want unambiguous rules on the use of AI.
These policies should state transparently which categories of data may be entered into particular tools and which approval processes apply to new software applications.
Improve technical detection and risk mitigation
Complete visibility is difficult to achieve through manual logging, so organisations must modernise their technical detection mechanisms.
CISOs should consider cloud access security broker services and advanced data loss prevention tools capable of identifying and analysing data transfers to known AI endpoints. The objective should be to block high-risk behaviour automatically.
Invest in continuous training and data literacy
The risks associated with modern AI systems are largely determined by how people use them. Technical barriers are insufficient if employees do not understand how the technology works.
Organisations should establish workforce-wide training programmes that go beyond basic instructions. Employees need to understand what may happen to their data after they submit a prompt.
Only by building strong data literacy across the workforce can organisations create a sustainable security culture in which innovation and risk awareness go hand in hand.
This article was first published on Computing Deutschland.