Is Your Confidential Data Safe in Free AI Tools?

CAG Executive Question of the Week

Can your leadership team explain what company information employees are entering into free AI tools—and what happens to that information afterwards?

Free AI tools are already part of the working day.

1. A finance manager asks one to summarise a monthly report.
2. An HR manager pastes in disciplinary notes to improve the wording.
3. A board secretary uploads a meeting transcript because producing the minutes manually will take another evening.

Nobody is trying to expose company information. Each employee is simply trying to work faster.

But before celebrating the time saved, management should ask a more uncomfortable question: what information has just left an approved business process?

Are Employees Using Personal or Unapproved AI Tools for Company Work?

Before answering, consider whether employees are using personal or free AI tools:

• Uploading financial reports, budgets, payroll records or confidential spreadsheets.
• Pasting board packs, management minutes, meeting transcripts or strategic plans into AI systems.
• Sharing customer information, contracts, proposals or commercially sensitive data.
• Entering employee records, CVs, disciplinary information or other personal data.
• Using free AI tools without knowing where information is processed, how long it is retained or who can access it.

If you answered yes to any of the first four questions, your organisation may already have an AI governance problem.

If the answer to the fifth is “we do not know”, the concern may be greater.

Management cannot govern activity it cannot see.

This does not mean every employee using free AI tools has caused a data breach. It means the organisation needs to establish what is being shared, why it is being shared and whether the use is permitted.

Free Does Not Automatically Mean Unsafe—and Paid Does Not Mean Approved

The debate is often reduced to a misleading choice: free AI tools are dangerous, while paid tools are safe.

That is not a reliable business rule.

A paid personal subscription may provide better features, larger usage limits and additional controls. It does not automatically give the employee’s organisation an appropriate contract, central administration, access management, retention policy or audit capability.

Equally, using a free service does not mean information is published on the internet for everyone to see.

The actual risk depends on the service, the account, its settings, the provider’s terms, the type of information and the organisation’s approved-use rules.

The better distinction is:

• Personal or organisation-managed
• Unapproved or approved
• Consumer service or properly contracted business service
• Permitted information or confidential information
• Uncontrolled output or human-validated output

A monthly subscription buys features. It does not create an AI governance framework.

What Happens to Data Entered Into Free AI Tools?

Information entered into an external AI service must be processed somewhere.

Depending on the provider and product, it may be stored for a period, reviewed for safety or support purposes, processed in another country, or used to improve services under particular settings and terms.

This is why executives should avoid sweeping claims about every platform.

ChatGPT, Microsoft Copilot, Gemini, Claude and specialist AI products do not all operate under identical terms. Free, personal, business, enterprise and API-based services may also have different protections.

For example, OpenAI explains how information from individual services may be used to improve its models and how users can control this setting.

OpenAI also states that its business offerings do not use organisational inputs and outputs to train its models by default.

Microsoft similarly publishes data, privacy and security commitments for Microsoft 365 Copilot.

These protections matter, but they do not remove the organisation’s responsibility to assess the tool and govern employee behaviour.

Executives should verify:

• What data the service collects
• Why the information is processed
• Whether prompts and files may be used for model improvement
• How long information is retained
• Where processing occurs
• Which subprocessors are involved
• Who can access the information
• Whether data can be deleted
• What contractual and security protections apply

The correct question is not merely, “Is this AI free?”

It is: “Is this specific use of this specific tool appropriate for this specific information?”

Free AI Tools and POPIA: The Issue Is Accountability

South Africa’s Protection of Personal Information Act does not prohibit organisations from using cloud platforms or external technology providers.

It does, however, require responsible parties to process personal information lawfully and apply appropriate safeguards.

An organisation should therefore not rely on an employee’s personal judgement alone when free AI tools are used for work involving personal information.

Personal Information Can Appear in Ordinary Documents

Employees may recognise an identity number or medical record as sensitive. They may not notice the personal information inside routine work such as:

• Meeting minutes containing employee names and opinions
• Spreadsheets containing salaries or performance information
• Customer complaints containing names and contact details
• CVs and interview notes
• Supplier records and banking details
• Emails describing disciplinary or health matters

Removing a person’s name may not be enough if the remaining information can still identify that person.

POPIA compliance requires more than adding a paragraph to an AI policy.

HR, legal, information security and operational leaders should verify the organisation’s position. Technical and legal controls must then be translated into instructions employees can apply during a busy working day.

Why Employees Turn to Free AI Tools

• It is easy to blame employees for creating what is often called “shadow AI”.
• That misses the business lesson.
• Employees usually adopt free AI tools because an existing workflow is slow, repetitive or frustrating.
• The board secretary does not want to bypass governance. She wants to avoid spending four hours producing minutes.
• The finance manager wants to meet a reporting deadline.
• The HR manager wants a clearer letter before the meeting starts.
• Prohibition alone will not remove that pressure.

If the approved process remains inefficient and no practical alternative is provided, employees will continue finding shortcuts.

That makes unapproved AI use a valuable management signal. It may reveal:

• Repetitive document production
• Poorly designed approval processes
• Manual reporting that should be simplified
• Weak Microsoft 365 skills
• Inconsistent templates and data structures
• A lack of approved AI options
• Training that focuses on prompting but ignores governance

Shadow AI is often not caused by reckless employees. It is caused by useful technology arriving before practical business rules.

Start With AI Workflow Discovery, Not Another Software Purchase

Buying a specialist software package may be appropriate, but software should not be the first response.

An AI Workflow Discovery process examines what employees are trying to accomplish, what information moves through the workflow, where delays occur and which controls must remain in place.

College Africa Group can assist you with AI consultancy.

Examine the Work Before Automating It

For each proposed AI use case, management should document:

1. The business objective.
2. The existing process and its bottlenecks.
3. The information required to complete the task.
4. The information classification and legal obligations.
5. The approved platform and account type.
6. The person responsible for validating the result.
7. The expected productivity or quality improvement.

This may reveal that the organisation does not need another platform.

Better templates, stronger Microsoft 365 skills or a redesigned approval process may solve much of the problem.

Strong workplace skills also help employees use AI more safely.

An employee who understands Advanced Microsoft Excel is better positioned to recognise confidential fields, structure information correctly and validate an AI-generated formula or analysis.

Employees who understand Microsoft Word, Outlook, Teams, PowerPoint and Loop are also less likely to copy information between disconnected systems simply to complete ordinary work.

In other cases, an organisation-managed AI service, ChatGPT for Business or Microsoft Copilot may support the approved workflow.

The decision should follow due diligence rather than product enthusiasm.

What Should an AI Acceptable-Use Standard Cover?

A practical standard should tell employees what to do, not merely warn them to “be careful”.

At minimum, it should define:

• Which AI tools and account types are approved
• Which information may never be entered
• When information must be anonymised or minimised
• Whether files, images, recordings and transcripts may be uploaded
• Which uses require manager, legal or information-security approval
• How AI-generated content must be checked
• How incidents and accidental disclosures must be reported
• Who owns the final decision and business output

The rules should be reinforced through scenario-based corporate AI training.

Employees need to practise deciding whether a payroll spreadsheet, customer complaint, board transcript or supplier contract is appropriate for a particular AI tool.

An unread policy will not change behaviour.

Practical judgement will.

Human Validation Still Matters in Secure Systems

Data protection is only one part of responsible AI adoption.

Even when an approved enterprise system protects information appropriately, an AI-generated summary can omit a decision, misstate a figure or assign an action to the wrong person.

A polished document is not necessarily an accurate record.

This is particularly important for financial reports, board minutes, HR correspondence and legal or compliance documents.

The accountable employee must compare the output with the original source, confirm material facts, check calculations and obtain the required approvals.

AI produces language. People remain responsible for decisions.

Is Your Confidential Data Safe in Free AI Tools? Ask Better Questions

Free AI tools can deliver genuine productivity benefits.

They can help employees organise ideas, improve non-sensitive wording, learn new concepts and explore better ways of working.

The objective should not be to frighten employees away from AI.

Nor should management assume that buying a paid product makes the risk disappear.

The objective is controlled, informed and useful adoption.

Start by identifying where free AI tools are already being used.

Review the business process. Classify the information. Assess the provider and account type. Provide an approved route. Train employees. Require human validation. Measure whether the workflow actually improves.

That is how a business moves from hidden experimentation to responsible business productivity.

Better Business Processes.
Stronger Workplace Skills.
Smarter AI.
Greater Business Productivity.

Take the Next Step

Do you know where employees are already using AI—and what business information they may be sharing?

Contact College Africa Group to discuss an AI Workflow Discovery session, practical AI governance training or the Microsoft 365 skills needed to create safer and more productive workflows.

Important: This article provides general business guidance and does not constitute legal advice. Your HR, legal, information-security and POPIA specialists should verify policies, contracts and processing arrangements for your organisation.

Frequently Asked Questions About Free AI Tools

Are free AI tools safe for confidential company information?

Businesses should not assume that free AI tools are approved for confidential information.

The level of protection depends on the provider, account type, settings, contractual terms, security controls and information involved.

Employees should use only tools and information categories approved by their organisation.

CAG’s AI in the Workplace training helps employees understand practical AI opportunities, risks and responsibilities.

Is a paid personal AI account safe for business data?

Not automatically.

A paid personal account may provide additional features, but it may not include organisational administration, an appropriate data-processing agreement, retention controls or audit capabilities.

Paying for an AI subscription does not make it an organisation-approved business service.

Organisations requiring a structured approach can explore CAG’s AI and ChatGPT consulting services.

Can employees use ChatGPT for company work?

That depends on the organisation’s AI policy, the ChatGPT service being used and the type of information involved.

Employees should use approved accounts, follow company information-handling rules and validate every important output.

Practical ChatGPT for Business training can help employees distinguish between appropriate productivity tasks and activities that may expose confidential information.

What company information should not be entered into unapproved AI tools?

Examples include payroll records, customer details, employee information, board packs, financial reports, contracts, passwords, confidential strategies, health information and legally privileged material.

Organisations should create clear examples based on their own workflows rather than expecting employees to interpret a vague instruction such as “do not share sensitive information”.

CAG’s Corporate AI Training can be adapted around the organisation’s policies, workflows and approved AI tools.

How do Microsoft 365 and Excel skills support safer AI use?

Strong Microsoft 365 skills help employees structure, analyse and manage information without unnecessarily transferring it to external tools.

Employees who understand formulas, Excel Tables, filters, PivotTables and data validation are also better equipped to check AI-generated calculations and conclusions.

CAG’s Advanced Microsoft Excel training develops the technical skills employees need to analyse information and validate AI-assisted work.

Better workplace skills reduce dependence on unapproved shortcuts.

Is Microsoft Copilot automatically safe for every company document?

No AI tool should be treated as automatically appropriate for every document.

Microsoft Copilot operates within Microsoft 365 and provides enterprise data-protection capabilities for qualifying organisational accounts.

However, the organisation must still manage permissions, information classification, licences and employee behaviour.

CAG’s Microsoft Copilot training covers practical productivity, responsible use, governance and human validation.

Does entering information into an AI tool automatically cause a POPIA breach?

Not necessarily.

The organisation must consider the type of information, purpose of processing, provider arrangements, security safeguards and possible cross-border processing.

Unauthorised disclosure or inadequate protection may create POPIA risk, but every situation requires proper assessment by the organisation’s legal, information-security and POPIA specialists.

Who is responsible for checking AI-generated business content?

The employee and organisation remain accountable for the final output.

AI-generated reports, meeting minutes, spreadsheet analysis, correspondence and recommendations should be checked against reliable source information and approved through normal business processes.

AI produces language.

People remain responsible for decisions.

Organisations that need help identifying suitable AI opportunities can start with CAG’s AI and ChatGPT consulting services.