Skip to main content

Our Thinking/AI & Governance

Shadow AI: the real risk is AI without rules

Your team is almost certainly using AI already, and that's a good thing. Here's what happens when it's used without any guardrails, and the simple steps that let your people keep the benefits safely.

By Digital Panda

· 5 minute read · Includes a 2 minute video

Watch first

Short on time? This covers the essentials in about two minutes.

Read the video transcript

Shadow AI

Hey you. I'm Shadow AI... and I'm coming for your business.

I whisper promises of instant results into your employees' ears.

When they're overwhelmed and just want to go home, I'm the unapproved, free tool they open in a hidden tab. They use me to write emails, fix spreadsheets, and summarise meetings.

Yes, they get the work done fast, without you ever knowing.

But I demand a heavy price.

I feed on your confidential client records. Your private financial data leaves your network and gets permanently absorbed into my public servers.

And you're none the wiser, because I take many forms.

Sometimes I'm a compromised app stealing your passwords. Sometimes I feed your staff a highly convincing lie, and they send it straight to your best customer.

And you can't stop me. You can't ban me. Your employees will just hide me better.

So, what are you going to do?

Give them secure, managed AI accounts? Limit what I can access? Tell them what they can and can't upload?

Go ahead. Try to shut me out. I dare you.

Digital Panda

Don't let Shadow AI take over your business.

One of the first steps to fix this is implementing an AI Acceptable Use Policy to help you set clear rules, protect sensitive data and give your staff a safer way to use AI.

Reach out today and we'll send you a template you can start from immediately.

In short

  • Shadow AI is staff using AI tools the business hasn't approved or set rules for.
  • It's rarely malicious. It's good people trying to work faster, without guidance.
  • The answer isn't a ban. It's an approved tool, a clear policy and some practical training.

The basics

What is shadow AI?

It's any AI used at work without the organisation knowing about it or agreeing how it should be used. Think "shadow IT", just with a much bigger appetite for data. It usually looks completely ordinary:

  • Pasting a customer complaint into a free chatbot to word the reply

  • Uploading donor or client records to "find the trends"

  • Running a board meeting recording through a transcription app

  • Installing a browser extension that summarises every email and web page

  • Letting an AI note-taker quietly join Teams or Zoom calls with clients

  • Checking a contract or HR letter in a personal AI account

The numbers

It's already happening, in most organisations

81%

of employees use AI tools their employer hasn't approved

Source: UpGuard, The State of Shadow AI

1 in 5

organisations studied had a data breach linked to shadow AI

Source: IBM, Cost of a Data Breach 2025

31%

of responding UK organisations have a policy to manage AI use

Source: IBM, Cost of a Data Breach 2025, UK findings

Why it happens

It's a gap, not a people problem

Shadow AI isn't a junior staff issue. UpGuard found executives were the most regular users of unapproved AI. The reasons are usually simple and very human:

  • No approved alternative

    If the business hasn't offered a safe tool, people find their own.

  • No clear rules

    Most staff have never been told which data is fine to use and which isn't.

  • It genuinely helps

    Once AI has saved someone an hour, a vague worry won't stop them.

  • Guidance hasn't kept up

    The tools change monthly. Most advice, where it exists, hasn't.

The risks

What can actually go wrong

A calm, honest look at the four risks we see most often. Each one is manageable once you know it's there.

Sensitive data leaves the building

Anything pasted into a free or personal AI tool goes to a third party's servers, where it may be stored or used to improve the service.

Real example

Within weeks of being allowed to use ChatGPT, Samsung engineers reportedly entered confidential source code and meeting notes into it. Samsung then banned generative AI tools altogether.

Data protection law still applies

UK GDPR has no exemption for AI. Personal data still needs a lawful basis, and the ICO expects a risk assessment for higher-risk uses.

Why it matters for you

Client records, donor lists, case notes and HR files are exactly what people are tempted to "just run through AI", and exactly what the ICO cares most about.

Confident answers that are wrong

AI can produce answers that sound authoritative but are simply made up. Whatever goes out still carries your name, not the tool's.

Real example

In June 2025 the High Court found 18 of 45 case citations in one set of legal submissions did not exist. The lawyers involved were referred to their regulators.

Tools nobody has checked

Some extensions can read everything on screen, and some note-takers record calls automatically. Without a review, nobody knows where that data ends up.

Watch out for

Apps that ask for access to your whole mailbox, calendar or file storage when all you wanted was a quick summary.

The cost

IBM found high levels of shadow AI added around $670,000 to the average breach, and those incidents were more likely to expose customers' personal information.

Why a ban doesn't work

45% of workers find a workaround when an app is blocked. Usage doesn't stop, it just moves to personal phones where you can't see it.

The more sustainable answer is to make the safe route the easy route: good tools, clear rules and the confidence to use both.

Source: UpGuard, The State of Shadow AI

The answer

Five steps to safer AI

No huge budget or team of specialists needed. For most SMEs and charities, it comes down to this.

Then review it every six months. The tools move quickly, and your approach should too.

  1. 01

    Find out what's already in use

    Ask your team openly, without blame, and check which apps are connected to your Microsoft 365 or Google accounts. You'll find good ideas worth keeping too.

  2. 02

    Give people an approved tool

    Business versions of mainstream AI tools generally come with stronger data protections and admin controls. A safe option removes most reasons to go elsewhere.

  3. 03

    Agree an AI acceptable use policy

    Short, in plain English, and clear on what's allowed, what isn't and who to ask. More on what goes into one below.

  4. 04

    Train for real life

    Practical habits, not an hour of slides: what's safe to paste, how to strip out personal details, and why every output gets a human check.

  5. 05

    Add light technical guardrails

    Control who can connect new apps, block unvetted browser extensions and use data loss prevention to keep sensitive files where they belong.

Step 03 in detail

What goes into an AI acceptable use policy

The best ones fit on a couple of pages and anyone can follow them. Start with a simple traffic-light rule for data:

Green

Fine to use

Public information, general research and drafting non-sensitive content.

Amber

Approved tools only

Internal documents, plans and processes, and only in your business AI accounts.

Red

Never, unless approved

Personal data, client or donor records, financial details and anything confidential.

The rest of the policy covers:

  • Which tools are approved, and how to request a new one
  • A person always checks and owns what's sent or published
  • When to tell clients or funders that AI was involved
  • No AI-only decisions about people's jobs, services or eligibility
  • Work accounts only, and no connecting apps without approval
  • A named owner, and a blame-free way to report mistakes

How we do it ourselves

Certified to ISO/IEC 42001

We use AI every day and help our clients do the same, so we hold ourselves to the standard we'd recommend. ISO/IEC 42001 is the first international standard for managing AI responsibly.

In practice, every AI tool we use is assessed before it's approved, we know exactly where data goes, and a person is always responsible for the result. It works alongside our other certifications, so security, quality and AI governance all pull in the same direction.

  • ISO 27001
  • ISO 9001
  • Cyber Essentials
ISO/IEC42001ASSESS RISKSET POLICYCONTROL DATAHUMAN OVERSIGHTREVIEW

Free AI usage review

Not sure what AI your team is already using?

We'll work alongside you to find out, with no jargon and no hard sell. Our free review gives you a clear picture and a sensible, proportionate next step.

What's included

  • A friendly conversation about how your team uses AI today
  • A check of which AI apps can reach your business data
  • A short, plain-English summary of your top priorities

Prefer to talk? Call 0121 227 0850

Keep reading

More from Our Thinking

See all articles

Written by

Digital Panda

Website, IT, security and AI for SMEs and charities. ISO 42001 certified.

Sources