AI in Business: The Data Risk Most Companies Have Not Yet Controlled

AI is already being used inside most businesses.

It may not have been formally approved. It may not appear on an IT roadmap. It may not have been discussed properly at leadership level. But staff are using AI tools to rewrite emails, summarise documents, analyse spreadsheets, prepare reports, create content, support customer service and reduce everyday administration.

That is not automatically a problem.

The problem is uncontrolled AI usage.

Many organisations are now in a position where AI is being used before the business has decided which tools are approved, what data can be entered, what should never be uploaded, and who is responsible for checking how AI is being used.

For business owners and leadership teams, this creates a practical risk. Customer data, employee information, financial records, supplier pricing, contracts, board reports and confidential operational detail may be copied into AI tools without anyone fully understanding the exposure.

The issue is not whether AI has value. It clearly does.

The issue is whether the business has control.

AI risk is not just a technology problem

AI data risk is often misunderstood because it is treated as an IT issue. In reality, it sits across operations, governance, compliance, commercial control and staff behaviour.

A business can have secure systems and still carry risk if an employee copies customer data into a personal AI account. A company can have good policies and still carry risk if nobody knows which AI tools staff are actually using. A leadership team can believe AI has not been adopted, while staff are already using it every day to save time.

This is why AI governance needs to be practical. It should not be a theoretical document that nobody reads. It should answer simple operational questions:

1. Which AI tools are being used?

2. Who is using them?

3. Are they personal accounts or business accounts?

4. What data is being entered?

5. Are files being uploaded?

6. Are outputs being checked before use?

7. Does the business have an approved process?

Without that visibility, the business is relying on individual judgement. That is not enough.

The hidden risk: staff trying to be efficient

Most AI risk does not start with bad intent.

It starts with someone trying to do their job faster.

An employee may paste a customer complaint into an AI tool to make the reply more professional. A manager may upload a spreadsheet to create a summary. A director may use AI to analyse a report before a meeting. A marketing person may create a custom assistant using company documents. A developer may connect an AI service into an internal workflow without a full review of what data is being sent.

Each individual action may appear reasonable. Together, they can create a serious governance gap.

The business may not know what information has been shared, where it has been entered, whether it is stored, whether it can be used for model improvement, whether the account belongs to the company, or whether the data can be removed later.

That is the real risk: not AI itself, but unmanaged AI behaviour.

Why personal AI accounts are a concern

One of the most common issues is staff using personal AI accounts for business activity.

This usually happens because personal tools are quick, familiar and already available. However, from a business perspective, this creates several problems.

The company may have no control over account settings. It may not know whether chat history, memory, file storage or model improvement settings are enabled. It may not be able to access, audit, manage or delete previous conversations. If the employee leaves, the company may have no clear way to recover or control the business information held inside that personal account.

For general, non-sensitive tasks, this may be a low-risk convenience. For customer, HR, finance, legal or commercially confidential information, it is a different matter.

Businesses should be especially careful where AI tools are being used with:

1. customer names, emails, addresses or order histories

2. complaints, disputes or service issues

3. employee records, sickness notes or disciplinary matters

4. payroll, invoices, aged debt or financial reports

5. supplier pricing, margin data or commercial terms

6. contracts, legal correspondence or board papers

7. internal processes, system exports or operational reports

These are not casual inputs. They are business data assets.

File uploads increase the risk

Copying a sentence into an AI tool is one thing. Uploading a full spreadsheet, PDF, contract or report is another.

Files often contain more information than the user realises. A spreadsheet may include hidden columns, notes, customer records, invoice values or historic data. A PDF may contain personal details, contract terms, metadata or commercially sensitive information. A report may include financial position, strategy, staffing issues or supplier details.

Staff may upload these files because AI makes analysis faster. The benefit is obvious. The risk is that the business may not have approved the tool, reviewed the account type, checked the settings or decided whether that data should be processed in that way.

Any business using AI should have a clear rule for file uploads. Staff need to know what is allowed, what requires approval, and what should never be uploaded into an unmanaged tool.

AI memory and retained context

Some AI tools can remember information, retain context or use previous conversations to improve future responses. This can make the tool more useful, but it can also allow company-specific knowledge to build up over time.

For personal use, memory can be helpful. For business use, it needs control.

If staff use AI repeatedly for company work, the tool may gradually build context around clients, suppliers, pricing, internal processes, writing style, decision-making preferences or operational issues. In the wrong account type, that creates a boundary problem between personal convenience and company-controlled information.

Businesses should check whether memory, chat history or retained context features are being used and whether they are appropriate for company data.

Custom AI assistants need approval

Custom AI assistants can be useful, especially where a business wants consistent answers, reusable instructions or support with repeated tasks.

They can also become unofficial internal systems.

A member of staff may create a custom assistant for proposals, HR, customer service, marketing or operations. To make it useful, they may upload company documents, policies, templates, price lists, client examples, scripts or internal procedures.

If this is done without review, the business may not know what information has been embedded, who can access it, whether it is still accurate, or whether it is being used in an approved environment.

Custom AI assistants should be treated like any other business system. They need ownership, access control, review and approval.

Connected apps can expose more than expected

AI tools are increasingly able to connect to email, calendars, cloud storage, documents, CRMs, project systems and other business applications.

This can be powerful. It can also widen the risk considerably.

If an AI tool is connected to a document store or email account, the question is no longer just what a user types into the chat. The question becomes what the tool can access, retrieve, summarise or reuse from connected systems.

Businesses should understand which connectors are enabled, what permissions have been granted, and whether those permissions match the intended use. Poorly controlled integrations can expose more information than the business expected.

API and embedded AI workflows

Some businesses are starting to use AI through APIs or software integrations. This can be a better route than staff manually copying data into public chat tools, because the workflow can be designed with more control.

However, API usage still needs governance.

The business should know what data is being sent, whether personal or confidential information is being minimised, whether requests and outputs are logged, where API keys are stored, who has access, and whether AI-generated outputs are reviewed before being used externally.

A controlled AI workflow can be valuable. An uncontrolled integration can simply automate the risk.

10 point AI data usage checklist

Every business using AI, formally or informally, should be able to answer these ten questions.

1. Which AI tools are staff using?

Identify the tools currently in use across the business, including chat assistants, writing tools, browser extensions, meeting note tools, image tools, document assistants and AI features inside existing software.

2. Are staff using personal or business accounts?

Check whether AI activity is happening through approved company accounts or unmanaged personal accounts.

3. Is model training or improvement switched on?

Understand whether prompts, responses, uploads or feedback may be used to improve AI models, and whether the company has control over those settings.

4. Are staff uploading files?

Review whether spreadsheets, PDFs, contracts, reports, customer exports or other business documents are being uploaded into AI tools.

5. Is sensitive business data being pasted into AI tools?

Check whether customer data, HR information, financial records, supplier pricing, legal documents or confidential reports are being entered into AI systems.

6. Is memory, chat history or retained context enabled?

Understand whether AI tools are retaining company-specific context, especially where staff are using personal accounts.

7. Have staff created custom AI assistants?

Identify whether custom assistants exist, what information has been added to them, who owns them and who can access them.

8. Are AI tools connected to business systems?

Review any connections to email, cloud storage, calendars, CRMs, document systems, finance tools or project platforms.

9. Is AI being used through APIs or internal software?

Check what data is being sent, how it is processed, how outputs are used, and whether the integration has been reviewed properly.

10. Does the business have an AI policy, staff training and approved workflows?

Ensure staff know which tools are approved, what data can be used, what must not be entered, when information should be anonymised, and when approval is required.

The aim is control, not obstruction

AI can create real commercial value. It can reduce administration, improve reporting, support customer service, speed up analysis, help with documentation and make better use of internal knowledge.

The answer is not to block AI.

The answer is to make sure it is used properly.

Businesses need a clear view of where AI is already being used, what data is being shared, which use cases are low risk, which require controls, and which should stop until an approved process is in place.

Uncontrolled AI usage creates exposure.

Controlled AI usage creates advantage.

Practical next step

An AI Data Usage Review gives business owners and leadership teams a clear understanding of current AI activity across the organisation.

It identifies which tools are being used, whether staff are using personal or business accounts, what data may be exposed, where higher-risk behaviours exist, and what practical controls should be introduced.

For many businesses, this is the sensible starting point. Before investing in larger AI projects, they need to understand what is already happening inside the organisation.

AI is moving quickly. Business control needs to catch up.