Shadow AI in police and fire services: the blue light data risk hiding in plain sight
For many blue light teams, shadow AI use is becoming a real concern. In this article, we discuss how you can bring AI back under control, enabling a safer, more confident way of working.
by Chris Whitehead VP of Government and Housing

Police and fire services hold some of the most sensitive data in the public sector. But to cope with workload, staff are increasingly turning to consumer tools such as Chat GPT and Claude to help with routine tasks – usually without approval. This unauthorised use of AI is known as shadow AI, and tends to surface most in finance, procurement and governance – areas where oversight and accountability matter most.
Below, we explain in more detail what shadow AI is, how widespread it has become and outline five practical steps to reduce the risk without slowing people down.
Shadow AI in 30 seconds
- Shadow AI is the use of AI tools at work without the organisation's approval or oversight.
- Microsoft research found 71% of UK employees have used unapproved consumer AI tools at work, and 22% have used them for finance-related tasks.
- Once sensitive data goes into an unapproved tool, you can no longer see where it is stored, how long it is kept or whether it is reused.
- Blanket bans rarely hold. What works is an approved tool people actually want to use, clear data boundaries and short, role-specific training.
- Ask every AI supplier where your data is processed as well as where it is stored, and whether it is ever used to train large language models.
What is shadow AI, exactly?
Shadow AI is the use of artificial intelligence at work that has not been approved, assessed or monitored by the organisation. Typical examples include pasting a budget report into a public chatbot to summarise it, using a free browser plug-in to rewrite tender wording, or uploading meeting notes to an AI note-taker linked to a personal account.
It is a close relative of shadow IT, the long-standing problem of staff using apps and cloud services that IT has not signed off. The difference lies in what happens to the data. A generative AI tool reads, processes and may retain whatever it is given, so a single prompt can move sensitive information outside the organisation's control in seconds. Many popular tools run on large language models (LLMs) hosted outside the UK, under consumer terms of service.
Shadow AI is rarely driven by malicious intent. More often, it occurs when employees use familiar AI tools to complete tasks more quickly, solve problems more effectively or reduce administrative effort.
How common is shadow AI in UK organisations?
Research commissioned by Microsoft UK and carried out by Censuswide in October 2025 surveyed 2,003 UK employees. It found that 71% had used unapproved consumer AI tools at work and 51% were doing so every week. Some 22% had used them for finance-related tasks, yet only 32% said they were concerned about the privacy of company or customer data entered into those tools.
Similar concerns are reflected at an organisational level. Research from Red Hat found that 83% of UK organisations reported experiencing a "shadow AI" problem, while SAP found that 68% of UK organisations reported employees using unapproved AI tools at least occasionally as part of their day-to-day work.
The findings show that shadow AI use is widespread, with employees embracing AI faster than many organisations can establish the governance, training and safeguards needed to support responsible use.
Where shadow AI shows up in police and fire services
Shadow AI rarely sits in one place. In blue light organisations it tends to appear in the support functions that keep frontline services running.
Finance: Summarising a budget monitoring report or variance analysis before a deadline, potentially exposing budget positions, supplier payments, reserves and savings plans.
Procurement: Rewriting a tender specification or tidying up evaluation commentary, potentially exposing commercially sensitive bids, scoring notes and contract values.
Governance: Turning rough notes into a report for a police and crime panel or fire authority, potentially exposing draft decisions, risk registers and information that has not yet entered the public domain.
HR and workforce: Drafting letters or summarising case notes, potentially exposing personal data relating to employees and staff.
Each of these scenarios appears relatively harmless. A finance colleague may summarise a document to save time, a procurement lead might refine tender language, or a governance officer could turn rough notes into a report ahead of sign-off. The work is completed more efficiently and, on the surface, no obvious harm has been done.
The issue is not the task itself, but the information involved. These teams routinely work with some of the most sensitive and highly scrutinised information within police and fire services. Financial decisions are subject to audit and public scrutiny, procurement processes must be transparent and defensible, and governance records provide a clear account of how decisions were made. As a result, the risks associated with shadow AI are often greatest in finance and procurement and in governance and risk functions.
What are the risks of shadow AI for sensitive public data?
The risk begins the moment information is shared. Once data has gone into an unapproved tool, the organisation can no longer see or control what happens to it. For police and fire services, that creates five specific exposures.
1. Loss of visibility. You cannot tell where the data went, who can access it or how long it will be kept.
2. Reuse and model training. Some consumer AI services use prompts to improve their models unless the user changes the settings. Sensitive content could be retained in ways you cannot audit.
3. Overseas processing. Data may be stored or processed outside the UK, making it more difficult to understand where information is held, how it is protected and which legal or regulatory requirements apply. This can create additional complexity around data governance, sovereignty and compliance.
4. Inaccurate outputs. AI tools can produce confident summaries that are wrong. If a flawed figure ends up in a board paper or tender evaluation, the organisation still owns the decision. 5
5. A weak audit trail. Work done in a personal AI account leaves no record in your systems. That makes it harder to answer audit queries or Freedom of Information requests about how a decision was made.
Why banning AI tools rarely works
The instinctive response is to block consumer AI sites and issue a policy. On its own, blocking tends to push use onto personal phones, where you have even less visibility. If people already use a chatbot at home, they will reach for it at work too.
Public sector guidance has moved in the same direction. The Artificial Intelligence Playbook for the UK Government, published by the Government Digital Service in February 2025, sets out 10 principles for using AI safely, effectively and securely. In policing, the Covenant for Using Artificial Intelligence in Policing was endorsed by all members of the National Police Chiefs' Council in September 2023. Both treat AI as something to adopt responsibly, with clear lines of accountability.
The real question is how to reduce the risk while keeping the productivity gains people are looking for.
How to reduce shadow AI risk: six steps for blue light leaders
Follow these six pratcial steps to begin reducing shadow AI use in your organisation:
1. Find out what is already being used
Start with an honest picture. Run a short, anonymous survey of finance, procurement and governance staff asking which AI tools they use and for which tasks. Pair it with your IT team's web and application logs to build an inventory of AI services in use. Make it clear there is no blame attached. If you lack the in-house capacity, managed cybersecurity services can help with discovery and ongoing monitoring.
2. Agree a short AI policy people can follow
Write a policy someone can apply in a hurry. One page is often enough: which tools are approved, what data must never go into an unapproved tool, and who to ask when in doubt. Align it with the AI Playbook principles and, for policing, the NPCC covenant. Our guide to AI governance covers how to build the wider framework around it.
3. Give people an approved tool that is as easy to use as the consumer version
This step does the most to cut shadow AI. If the approved route feels slower or harder than a consumer tool, people will go back to what they know. Choose AI that sits inside the systems your teams already use for finance, spend and governance, so the safe option is also the quickest one.
4. Set firm data boundaries with every AI supplier
General assurances of security leave too much unsaid. Police and fire services need clarity on sovereign data handling, so that sensitive information remains under UK jurisdiction and is processed entirely within the UK. Data stored in a UK data centre can still be processed elsewhere, so ask about processing as well as storage. The baseline is simple: sensitive data stays in the UK, is processed in the UK and never trains external large language models.
5. Keep training light and role-specific
Training matters, but it should never become a barrier. Short sessions built around real tasks, such as summarising a budget report or drafting a tender clarification, beat generic e-learning. Show people what good use looks like in their own role and they are far more likely to stay on the approved route.
6. Make someone accountable
In most police forces and fire and rescue services, responsibility for information risk will sit with the Senior Information Risk Owner (SIRO), working closely with the Data Protection Officer and other governance leads. Regular reporting and oversight help ensure risks associated with AI use are understood, monitored and managed appropriately.
Questions to ask an AI supplier about data sovereignty
Use these questions during procurement exercises, contract renewals and supplier reviews. A responsible supplier should be able to answer each one clearly and in writing.
Where is our data stored, and where is it processed?
Storage and processing are not always the same thing. Data may be stored in the UK while being processed elsewhere.
Is our data ever used to train or improve large language models, either yours or those provided by third parties?
Training activities can introduce additional risks if sensitive information is used to improve large language models beyond your control.
How long are prompts, files and outputs retained, and can retention periods be configured?
Retention policies determine how long information remains accessible and how long any potential exposure lasts.
Which subprocessors can access our data, and where are they based?
Third-party suppliers and subcontractors may introduce additional data sovereignty and compliance considerations.
Which independent certifications support your AI governance and information security practices?
Standards such as ISO/IEC 42001, ISO 27001 and ISO 9001 provide independent evidence that governance and security controls are in place.
What audit logs and usage records are available?
Organisations need clear audit trails to support investigations, governance reviews, audit requirements and Freedom of Information requests.
Can access be restricted by role, team or department?
Sensitive financial, procurement and governance information should only be accessible to authorised users.
How OneAdvanced AI supports safe AI use in blue light teams
OneAdvanced AI is designed to streamline workflows and support decision-making, with sector-specific AI agents and configurable privacy controls built in. All data is hosted, processed and controlled exclusively by OneAdvanced within the UK, with oversight aligned to public sector requirements
At its core are the OneAdvanced Models, which combine a large language model with sector-specific small language models (SLMs). In practice, that means answers shaped by the language and processes of the sector you work in. Alongside these, AI agents support a range of tasks and processes. They can be embedded directly in OneAdvanced products or deployed as standalone capabilities, using generative AI, natural language processing and machine learning. See how this applies to public bodies on our OneAdvanced AI for government page.
The experience is intuitive and access-controlled, so teams get value without becoming AI specialists. A key part of this is Private Spaces: secure environments for uploading and working with files that keep sensitive information contained while it stays usable. Private Spaces are built on the Model Context Protocol (MCP), an open standard for connecting AI models to data and tools, which supports context-aware AI while prioritising security and scalability. For finance, spend and governance teams, this builds good practice into everyday work rather than correcting behaviour afterwards.
For assurance, OneAdvanced holds ISO 42001 certification for AI governance, and our Trust Centre sets out how we protect customer data.
Our view: making AI safe by making it usable
Shadow AI tells you something useful: people want to work faster and are reaching for whatever helps.
For police and fire services, the challenge is to harness that momentum without losing control of the data and decisions that underpin public trust. That means making sure that a supplier's promise of security includes UK-based processing, clear data boundaries and explicit assurances that sensitive information will not train external models or be handled overseas.
The organisations that handle this well will put secure, intuitive tools in people's hands, keep training practical and design governance so it fits naturally into daily work. When that happens, teams can use AI with confidence every day, and leaders can account for how it is used.
Don’t miss!
Our IQ for Government page features finance, spend, governance and workforce solutions tailor-made for police and fire services in the UK – all underpinned by sovereign AI.
FAQs
About the author
Chris Whitehead
VP of Government and Housing
With nearly 25 years at the company, Chris leads a team supporting local authorities and housing providers. Known for his deep sector insight, Chris builds strong partnerships across public and private sectors to deliver innovative digital solutions that improve service delivery and citizen outcomes.
Contact our sales and support teams. We're here to help.
Speak to our expert consultants for personalised advice and recommendations or to book a demo.
Call us on
0330 343 4000Please enter your details, and our team will contact you shortly.
All fields are required
From simple case logging through to live chat, find the solution you need, faster.
Support centre