Shadow AI Examples: The Most Common Unsanctioned AI Tools in UK Workplaces

The most common unsanctioned AI tools found in UK workplaces are ChatGPT (free tier), Grammarly, Otter.ai, Notion AI, Perplexity, Claude.ai, Copy.ai, Midjourney, GitHub Copilot on personal accounts, and browser-based AI writing extensions. Meeting transcription tools are particularly prevalent — employees use them to do their jobs faster without understanding that recordings of client calls may be processed on servers outside the UK. Most shadow AI gets in not through deliberate circumvention but through employees trying to be productive.

How shadow AI gets into your business

Shadow AI does not usually arrive through deliberate policy circumvention. It arrives through convenience. An employee discovers a tool that makes a tedious task faster, tries it on their personal email, finds it works well, and starts using it for work. They do not think about the data handling implications. They do not raise a ticket with IT. They just get on with their job.

The pathways are predictable: browser extensions (no installation approval required, invisible to IT asset inventory), "Sign in with Google" on SaaS tools (records in the OAuth log but rarely monitored), personal mobile apps (entirely invisible to corporate controls), and copy-paste workflows where employees manually move work data into consumer AI tools.

The result is a gap between what IT thinks is running and what is actually running — and that gap is where data risk accumulates.

The 10 most common shadow AI tools found in UK workplaces

1. ChatGPT (free and Plus tiers)

The most prevalent by far. Employees use it for drafting, summarising, researching, and generating content. The free tier's data handling terms permit OpenAI to use inputs for model training by default. Employees routinely paste client names, contract language, financial data, and internal communications into it without considering the implications.

2. Grammarly

So common it is almost invisible — many employees do not consider it an "AI tool." The browser extension reads everything typed in the browser, including email content, form fields, and documents. The free and premium tiers have different data handling terms to the enterprise product. Often found installed on company devices without IT knowledge.

3. Otter.ai and similar meeting transcription tools

Employees invite transcription bots to Zoom or Teams calls to get automatic notes. The recordings and transcripts — which may contain confidential client discussions — are stored on third-party servers. Clients are often not informed that calls are being transcribed by a third-party service. This creates potential GDPR issues and professional conduct risks in regulated sectors.

4. Notion AI

Notion is increasingly used as a company wiki and project management tool. Notion AI sits inside it. If employees are using personal Notion accounts for work notes (common), or if the company Notion account does not have an enterprise agreement with appropriate data handling terms, Notion AI represents a significant shadow risk.

5. Perplexity

Used as an AI-powered research tool. Employees use it to research competitors, summarise industry reports, and answer factual questions. Lower-risk than tools that process personal data, but employees sometimes paste internal strategy documents or client briefs into it as context for their questions.

6. Claude.ai (consumer tier)

Anthropic's Claude is increasingly used as an alternative to ChatGPT. The consumer tier (claude.ai) is distinct from the API and enterprise offerings. Usage is growing rapidly and is almost universally discovered via OAuth audit rather than IT inventory.

7. Copy.ai and Jasper

Marketing and sales teams use these for content generation. Often adopted at team level without central IT involvement. Typically connected via work email OAuth, meaning they may have access to email and calendar data even if the employee only intended to use them for copywriting.

8. Midjourney and image generation tools

Used by marketing teams for image creation. Midjourney operates via Discord, which means company data shared with it flows through Discord's servers — an additional third-party processor that most businesses have not reviewed. Intellectual property ownership of generated images is also an unresolved legal question.

9. GitHub Copilot (personal accounts)

Developers use Copilot on personal GitHub accounts to access it without going through an IT approval process. Personal accounts use code inputs for model training; enterprise accounts do not. Developers pasting proprietary code into personal Copilot sessions may be sharing trade secrets with Microsoft's training dataset.

10. Browser AI extensions

A broad category covering Merlin, Monica, Sider, and dozens of similar tools that add AI functionality to the browser. Most request broad permissions (read all your browsing data, read and change data on all sites). Invisible to OAuth audit because they use extension permissions rather than OAuth flows. Only discoverable via employee survey or device management tools.

What data these tools can access

The risk is not always obvious from the tool's primary function. A writing assistant seems harmless until you realise the browser extension is reading every field the employee types in — including login forms, CRM notes, and email drafts. A meeting transcription tool seems like a productivity win until a client discovers their confidential conversation is stored on a server in Virginia.

The highest-risk access types to look for: email read access (common in tools that help draft replies), calendar access (reveals meeting participants and subjects), file access (Drive or SharePoint), and browser extension permissions (broadest of all).

Why blocking alone does not work

Blocking specific tools at the network level or via Google/Microsoft admin controls is a valid tactical response to high-risk tools. But it is not a governance strategy. Employees who want to use a tool will find alternatives — switch to a competitor tool, use personal mobile data, or use a personal device. The underlying need does not go away.

The more effective approach is to understand why employees are using shadow AI (what need it meets), provide sanctioned alternatives that meet the same need safely, and pair that with a clear policy that explains the why behind restrictions. Governance that makes sense to employees gets followed. Governance that feels arbitrary gets worked around.

How Governably detects shadow AI

Governably scans your Google Workspace and Microsoft 365 OAuth grants against a database of known AI tools, flags each by risk tier, and surfaces the results as a prioritised remediation list. Tools that appear in your OAuth log but not in your approved tools list are automatically flagged as shadow AI candidates — giving you a starting point for investigation rather than a blank canvas.

Frequently asked questions

What is shadow AI?

Shadow AI refers to AI tools that employees use at work without IT knowledge or approval. It is the AI equivalent of shadow IT — tools that exist outside the official technology stack and therefore outside the organisation's data governance and acceptable use policies.

What are the most common shadow AI tools in UK workplaces?

The most commonly found unsanctioned AI tools in UK business environments are: ChatGPT (free tier), Grammarly, Otter.ai, Notion AI, Perplexity, Claude.ai, Copy.ai, Midjourney, GitHub Copilot on personal accounts, and browser-based AI writing extensions.

Is shadow AI a GDPR risk?

Yes. If an employee shares personal data with an unsanctioned AI tool, the business may be in breach of UK GDPR. The lawful basis for processing personal data almost certainly does not extend to sharing it with a third-party AI provider whose data handling terms have not been reviewed.

Why does blocking shadow AI tools not work?

Blocking specific tools creates a whack-a-mole problem — employees switch to alternatives or use personal devices. The more effective approach is to provide sanctioned alternatives that meet the same need safely, paired with a clear policy explaining the reasons behind any restrictions.