020 3802 6525

[email protected]
astec logo
Contact Us
Book Your IT Review
Book Your IT Review

Shadow AI: The AI Tools Your Staff Use That Your IT Partner Can’t See

A practical guide for London and UK SMEs in any sector: what shadow AI is, what it actually risks, how to find out which tools your team already uses, and the one-page AI policy that fixes most of it.

By Sanish Satheesan CISSP, CIO at Astec IT · August 2026

Someone in your business used an AI tool this week that you don’t know about. Shadow AI — staff using unapproved AI tools for work — is now the normal state of affairs in UK companies, not the exception, whatever your business does. It looks different in an estate agency than it does in an insurance brokerage, but the shape of the risk is identical. This guide is written for owners, MDs and finance leads at London and UK businesses between 10 and 100 staff: what shadow AI is, the five risks that matter commercially, how to discover what’s already in use, what belongs in a one-page AI policy, and what your managed IT support company should already be doing about it.

IT Review

Book Your Cyber Security Review

No obligation. Speak with a dedicated cybersecurity specialist and receive practical recommendations tailored to your business requirements.
Book Your IT Review
020 3802 6525

What is shadow AI?

Shadow AI is any artificial intelligence tool used for work without the business approving, securing or even knowing about it. In practice that means a free chatbot, transcription tool, image generator or AI browser extension that a member of staff signed up for with their work email. It is the AI version of shadow IT.

Nobody is being reckless. Your team is trying to get through the work faster. A salesperson pastes a client email chain into a chatbot to draft a reply. Someone in finance drops a spreadsheet into a free analysis tool. An account manager runs a client call through an AI notetaker that quietly joined the meeting.

Each of those is a rational decision by a busy person. Collectively they mean your commercial data now sits in accounts you don’t control, can’t audit and can’t switch off.

The scale isn’t marginal. Research by Censuswide for Microsoft, published in October 2025, found 71% of UK employees have used unapproved consumer AI tools at work, and 51% do so every week. Nearly half used them for drafting workplace communications and 22% for finance-related tasks. Only 32% said they were concerned about data privacy. More recent research published by KnowBe4 in July 2026 found 55% of UK employees admit to using unapproved tools, and one in ten had knowingly entered sensitive information into an AI platform anyway.

What does shadow AI look like, sector by sector?

The tools and the excuse for using them change by sector, but the pattern doesn’t: someone pastes sensitive information into a consumer AI account to save time. What counts as sensitive is different in a law firm than a hotel, but the exposure is the same shape everywhere.
We work across real estate, financial services, insurance, hospitality and professional services, and shadow AI shows up in every one of them — just wearing different clothes.
SectorWhere it shows upWhat’s actually at risk
Real estateAgents drafting vendor correspondence or property particulars through a free AI toolVendor and buyer financial details, ID documents used for anti-money-laundering checks
Financial servicesAnalysts summarising client portfolios or drafting client emails with a consumer chatbotInvestment positions, personal financial data, regulated client records
InsuranceBrokers drafting policy wording or claims correspondence with a free AI assistantClaims history, medical detail, underwriting data
HospitalityReservations and events staff using AI to draft guest communicationsGuest personal data, payment details, dietary and medical notes
Professional servicesConsultants and account managers running an AI notetaker on client callsClient-confidential discussion, contract terms, information covered by an NDA
None of this is about the sector being careless. It’s about every sector handling data it can’t afford to lose control of, and AI tools making it easier than ever to move that data somewhere nobody signed off on.

What's in this guide

  1. What is shadow AI?
  2. What shadow AI looks like, sector by sector
  3. Why shadow AI is a risk for a small business
  4. What shadow AI actually costs a UK business
  5. How to find out which AI tools your staff use
  6. What an AI policy for a small business should include
  7. Who should own AI risk in your business
  8. Is Microsoft 365 Copilot safer than free AI tools?
  9. IT support for London and UK SMEs
  10. Frequently asked questions

Share This Article

Why is shadow AI a risk for a small business?

Because company data leaves your control and you have no record of it. Client details, financial figures and contracts get pasted into consumer accounts you cannot audit, restrict, log or switch off when someone leaves. You also lose the ability to answer a client, an auditor or the ICO when they ask where your data went.
The government’s own figures show the gap clearly. The Cyber Security Breaches Survey 2025/2026, published on 30 April 2026, found that 31% of UK businesses were using AI, adopting it or actively considering it — but of those, only around a quarter had any cyber security practices or processes in place to manage the risks from it.

Roughly three in four UK businesses engaging with AI have no process for managing its risks. The same survey found 43% of UK businesses reported a breach or attack in the previous 12 months, rising to 46% of small businesses and 65% of medium-sized ones, with phishing the most prevalent attack by far at 38%.

Image Alt

Your data leaves and you can’t get it back

Once a client list, a draft contract or a set of management accounts has been pasted into a consumer AI account, you have no delete button that you control. Free and personal-tier services often use inputs to improve their models by default. You cannot un-share information, and you cannot tell a client with confidence that their data never left your systems.
Image Alt

You can’t prove what happened

The ICO published its early view on agentic AI in January 2026 and the message for businesses was direct: the organisation remains responsible for data protection compliance, whatever tool an employee chose. The regulator flagged scope creep — systems processing more personal information than the task needs — and confusion over who is controller and who is processor across the AI supply chain. Neither of those questions has a good answer if the tool was never approved.

Image Alt

AI accounts sit outside your security controls

An account created with a work email and a personal password is invisible to your identity controls. It typically has no multi-factor authentication, no logging you can review, no conditional access rules, and — the one that matters most — no offboarding. When that employee leaves, the account and everything in it leaves with them. Proper account governance starts with knowing which accounts exist.
Image Alt

Agentic AI doesn’t just answer, it acts

The newer wave of AI tools can take actions on your behalf: send emails, move files, update records, connect to other systems. In May 2026 the NCSC published guidance on adopting agentic AI carefully, warning that these systems often get overly broad access to data and tools, and that they act faster than anyone can review. Its position, paraphrased plainly: if you cannot understand, monitor or contain what an agent does, it isn’t ready to deploy. An agent nobody approved is by definition unmonitored.
Image Alt

It quietly breaks your Cyber Essentials scope

This is the one most businesses miss. The Cyber Essentials requirements that took effect on 27 April 2026 (v3.3) introduced a formal definition of a cloud service — an on-demand, scalable service hosted on shared infrastructure and accessible via the internet — and confirmed that cloud services cannot be excluded from scope. Multi-factor authentication is mandatory on cloud services where it is available.

Read that against a free AI tool your team signs into with work credentials and fills with company data. It is very hard to argue out of scope. If you hold or want Cyber Essentials certification — and increasingly your clients and insurers will ask — an undocumented AI tool is a live problem, not a theoretical one. Confirm the scope with your certification body, then get the list of tools in front of them.

What does shadow AI actually cost a UK business?

The costs are rarely a fine. They are lost deals when a client security questionnaire can’t be answered, rework when AI output goes out wrong, duplicate subscriptions on expenses, and the time your own people spend untangling it. The expensive version is a data incident you find out about from someone else.
Most UK SMEs will feel this commercially before they feel it legally. Corporate clients now send security questionnaires as standard, cyber insurance renewals ask about AI use, and lenders and regulators ask similar questions in their own paperwork — whatever sector sends the form. “Which AI tools process our data, and under what agreement?” is on all of them. If your honest answer is that you don’t know, that is a procurement problem in a live sales process.

There’s a quieter cost too. Where there’s no approved option, people improvise — 28% of employees in the Microsoft research said their company provides no approved AI alternative. Banning AI outright doesn’t remove the risk. It just moves the usage to personal devices where you definitely can’t see it.

How do I find out which AI tools my staff are already using?

Start with your Microsoft 365 or Google Workspace sign-in logs and enterprise application list, which show third-party apps staff have connected using work accounts. Then check browser extensions on managed devices, review card and expense claims for AI subscriptions, and ask your team directly without blame.

1. Pull the connected applications list. In Microsoft 365, the enterprise applications and sign-in logs in Entra ID show every third-party service your staff have granted access to with a work identity. This single report surprises most owners.

2. Audit browser extensions and installed apps. AI assistants and notetakers usually arrive as extensions. Device management on managed endpoints will list them; this is standard IT infrastructure hygiene, not a special project.

3. Check expenses and card statements. Look for AI subscriptions on personal cards claimed back. Finance often finds shadow AI before IT does.

4. Look at meeting invitations. AI notetakers join calls as participants. If a bot has been attending client meetings, the invite history will show it — and so will your recording consent obligations.

5. Ask the team, blame-free. Tell people you want to make good tools available, not confiscate them. You will get an honest list in a morning. Lead with a ban and you’ll get silence.

6. Write the list down and keep it. A living register of approved and discovered tools is what turns a security questionnaire from a scramble into a five-minute job.

What should an AI policy for a small business include?

An approved list of tools, a clear statement of what data must never be entered, a route for staff to request new tools, a rule that AI accounts use work identities with multi-factor authentication, human review before AI output is sent externally, and a named person who owns the policy. One page is enough.

Long policies don’t get read. This is the version that works for a 10 to 100 staff business — use it as a checklist.

  • An approved tool list — two or three named tools people may use, and permission to use them properly.
  • A never-paste list — client personal data, employee records, card and bank details, passwords, contracts under NDA, anything special category.
  • A request route — who to ask for a new tool, and a commitment to answer within days rather than months.
  • Work identity only — AI accounts created through your single sign-on with multi-factor authentication, never a personal email and password.
  • A human check before anything goes out — AI drafts client-facing work; a person signs it off.
  • Disclosure rules for recording — tell people when an AI notetaker is on the call, every time.
  • Leaver process included — AI accounts are removed on the same checklist as email and file access.
  • A named owner — one person accountable for the policy, and a review date in the diary.
  • Training, not just a document — fifteen minutes of staff security training beats a policy nobody opens.
  • A recovery position — know what you’d do if data did go somewhere it shouldn’t, and make sure your backups and recovery plan are tested.

Who should own AI risk in your business?

Someone should, the same way someone owns your insurance renewal or your GDPR compliance. In a 10–100 staff business that’s usually an operations or finance lead who makes it part of onboarding and IT policy, working with whoever manages your technology, whether that’s an in-house person or an outsourced IT partner. What matters is that somebody is accountable, not the org chart.
Treat AI the way you’d treat any other business risk with a paper trail: name an owner, write it down, and revisit it on a schedule. Here’s the practical difference between a business where nobody owns this and one where somebody does.
AreaNobody owns itSomebody owns it
DiscoveryNobody knows which AI tools are connected to company accounts.Connected apps reviewed on a schedule and written down.
Raising the subjectComes up only after something goes wrong.Reviewed before it becomes a problem.
Access controlPersonal passwords, MFA inconsistent.Work identities with multi-factor authentication enforced.
Staff who leaveEmail disabled, third-party AI accounts forgotten.AI accounts included in the leaver checklist.
Compliance questionsAnswered from scratch, under time pressure.Answered from an existing tool register.
Cyber Essentials scopeCertified once, scope never revisited.Scope reviewed as cloud and AI tools are added.
Forward planningNobody revisits it until the next incident.Reviewed on a set schedule, AI included.

If IT is outsourced, this kind of discovery, access control and leaver management is exactly the work a competent IT support company should already be doing without being asked. We build it into a technology review twice a year for every client, alongside dedicated primary and secondary engineers and ISO 27001:2013 and ISO 9001:2015 accreditation for how we handle data ourselves. But the ownership question comes first — decide who’s accountable in your business before you decide who helps them do it.

Is Microsoft 365 Copilot safer than free AI tools?

For business data, generally yes, because it operates inside your Microsoft 365 tenant under your existing identity and permission controls rather than in a personal consumer account. The catch is that it inherits your permissions, so if files are over-shared internally, Copilot surfaces them faster than anyone expected.

Giving people a good approved option is the most effective control available, and it’s cheaper than dealing with the alternative. But “we bought Copilot” is not a governance strategy on its own.

Before you switch it on, tidy your permissions. Most SME tenants have folders shared with “everyone in the organisation” that nobody has reviewed in years — salary data, board papers, HR files. Copilot won’t break your permissions, it will obey them. That work — sorting sharing, sensitivity labelling and secure remote access — is normal cloud housekeeping, and it should be done before rollout, not after the first awkward discovery. It’s also worth checking proactive monitoring covers the new activity.

IT support for London and UK SMEs

Astec IT provides managed IT support in London and across the UK for businesses with 10 to 100 staff. London is our primary market, with clients also in Hertfordshire, Kent, Surrey, Leicester, Liverpool and Manchester. Support is remote-first, with on-site visits where they’re genuinely needed.

Does your IT support company need to be local?

For AI and cloud governance, no. This work is identity, permissions and policy — all delivered remotely, and resolved faster that way. Being local matters for on-site incidents, office moves and hardware. What matters more is whether the company has real coverage in your area or subcontracts every visit to whoever is nearest.

Why London SMEs switch IT support company

The pattern is consistent: not a disaster, but a slow realisation that nobody is thinking ahead on their behalf. AI is the current example. A London business with 40 staff and a client base that sends security questionnaires needs an IT partner that raised AI governance last year, not one that waits to be asked. Our client case studies follow that shape, and we’re happy to put you in touch with a client directly.

London • Greater London • Hertfordshire • Kent • Surrey • Leicester • Liverpool • Manchester

Frequently asked questions about shadow AI

What is shadow AI?
Shadow AI is any artificial intelligence tool used for work without the business approving, securing or even knowing about it. It usually means a free chatbot, transcription tool, image generator or AI browser extension that a member of staff signed up for with their work email. It is the AI version of shadow IT.
Why is shadow AI a risk for a small business?
Because company data leaves your control and you have no record of it. Client details, financial figures and contracts get pasted into consumer accounts you cannot audit, restrict, log or switch off when someone leaves. You also lose the ability to answer a client, an auditor or the ICO when they ask where your data went.
How do I find out which AI tools my staff are already using?
Start with your Microsoft 365 or Google Workspace sign-in logs and enterprise application list, which show third-party apps staff have connected using work accounts. Then check browser extensions on managed devices, review card and expense claims for AI subscriptions, and ask your team directly without blame.
Does shadow AI break UK GDPR?
It can. If staff put personal data into an AI tool you have not assessed, you may be processing that data without a lawful basis, without a data processing agreement and without knowing where it is stored. The ICO’s position is that the organisation remains responsible for compliance regardless of which tool an employee chose.
Does shadow AI affect Cyber Essentials certification?
Potentially, yes. The April 2026 Cyber Essentials requirements define a cloud service broadly and state that cloud services cannot be excluded from scope, with multi-factor authentication mandatory where it is available. An AI tool holding company data, accessed with work credentials, is hard to argue out of scope.
What should an AI policy for a small business include?
An approved list of tools, a clear statement of what data must never be entered, a route for staff to request new tools, a rule that AI accounts use work identities with multi-factor authentication, human review before AI output is sent externally, and a named person who owns the policy. One page is enough.
Is Microsoft 365 Copilot safer than a free AI chatbot for work data?
For business data, generally yes, because it operates inside your Microsoft 365 tenant under your existing identity and permission controls rather than in a personal consumer account. The catch is that it inherits your permissions, so if your files are over-shared internally, Copilot will surface them faster than anyone expected.
Who should manage AI risk in a small business?
Someone accountable for it, whether that’s an internal operations or finance lead or your outsourced IT partner. Discovering which cloud and AI services staff use, controlling access, removing accounts when people leave and advising on an approved toolset can sit with either, as long as one of them actually owns it.
Do you provide IT support for London businesses?
Yes. Astec IT is a London-based managed IT support company and London is our primary market. We support businesses with 10 to 100 staff across London and Greater London, plus Hertfordshire, Kent, Surrey, Leicester, Liverpool and Manchester, with remote-first support and on-site visits where they are needed.
Does my IT support company need to be local to my office?
For work like AI and cloud governance, no, because it is delivered remotely and resolved faster that way. Local matters for on-site incidents, office moves and hardware. Choose an IT support company with genuine coverage in your area rather than one that subcontracts every visit.
What is the best IT support option for a London SME using AI tools?
Fully managed IT support on a fixed monthly fee, including identity and cloud application management. For a London SME between 10 and 100 staff that covers the help desk, proactive monitoring, security fundamentals and AI governance for less than the cost of one senior IT hire.

What to do next

You don’t need an AI strategy this quarter. You need to know what’s already connected to your business, someone named as accountable for it, and one page telling your team what’s allowed. All three are a week’s work at most, in any sector.

Start with the question, not the vendor: which third-party AI and cloud applications are our staff signed into with work accounts, right now? Whoever manages your technology — an internal lead or an outsourced IT partner — should be able to produce that list without a special project. If nobody can, that’s the gap to close first. If you want more background, our resource library for UK SMEs covers the related ground on security and switching.

NO COMMITMENT

Not sure who owns this yet? We’ll run the discovery for you.

An AI and cloud application review of your Microsoft 365 tenant shows exactly which tools your staff are connected to, whoever ends up managing it from there. You keep the report either way.

About the author

Sanish Satheesan CISSP is Chief Information Officer at Astec IT. He has worked in IT for more than 15 years, including 11 years at Netstar, where he rose from service engineer to professional services manager, and two and a half years as Information Security Manager at CRU. Since 2021 he has held the ISC2 Certified Information Systems Security Professional (CISSP) certification, and he studied at Kingston University. He writes regularly about cyber resilience, incident response and IT audits for UK SMEs.

Connect with Sanish on LinkedIn

Related reading

Sources: Microsoft UK / Censuswide, shadow AI research (2,003 UK employees, October 2025) · KnowBe4 research via Security Journal UK (July 2026) · DSIT Cyber Security Breaches Survey 2025/2026 (30 April 2026) ·ICO Tech Futures: Agentic AI (January 2026) · NCSC, Thinking carefully before adopting agentic AI (May 2026) · IASME, Cyber Essentials April 2026 update (v3.3).   

    Book Your IT Review