Some businesses are still debating whether they should allow artificial intelligence in the workplace.
Meanwhile, their employees are already using it.
They are opening ChatGPT, Claude, DeepSeek and other public tools to rewrite emails, summarise documents, brainstorm responses and make everyday tasks easier.
In many cases, this use is happening with good intentions.
Employees are not necessarily trying to bypass policy or expose company information. They are trying to work more efficiently using tools that are readily available.
The problem is that the business may have no visibility over what information is being entered, where it is going or how it may be used.
This was one of the most important points raised by Michael Brook during his presentation at our recent Geelong Cyber Sip event.
Michael’s session explored Microsoft Copilot and the growing role of AI agents inside businesses. While he demonstrated several ways these tools can improve productivity, he also made it clear that AI adoption cannot be separated from data security.
Businesses need to make AI useful.
They also need to make it safe.
That balance is one of the reasons we run Cyber Sip events.
These events are designed to give local organisations access to current information from people working directly with these technologies.
AI is changing quickly, and many businesses are trying to make decisions without a clear picture of what their employees are already doing.
Bringing these conversations into the open is an important first step.
Banning AI does not mean it is not being used
A business may decide it is not ready to introduce AI.
That decision does not remove AI from the workplace.
Employees have access to these tools on their phones, home computers and web browsers. Many are already familiar with them through study, personal use or previous employment.
Michael noted that younger people entering the workforce are likely to see generative AI as a normal part of how they work.
They use it to explore ideas, find information and improve what they produce.
When they join an organisation, they will continue looking for those advantages.
A policy that simply says “do not use AI” may not be enough.
It can push the activity out of sight.
The employee still uses the tool, but the organisation loses any opportunity to guide, secure or monitor that use.
This creates what is often called shadow AI: artificial intelligence being used without formal approval, visibility or controls.
The risk begins with ordinary information
The information entered into a public AI tool may not look sensitive at first.
Someone copies an email into the system and asks for a more professional response.
A staff member pastes part of a customer document and asks for a summary.
A manager uploads notes and asks the tool to prepare a report.
The employee sees a simple productivity shortcut.
The business needs to consider what may be contained inside that information.
The email could include a customer’s name, contact details or financial information.
The document may contain intellectual property.
The notes might refer to staff performance, legal matters or an upcoming commercial decision.
Once that information has been entered into an external platform, the business may no longer have the same control over it.
Different AI services have different terms, privacy settings and approaches to model training.
Employees are unlikely to review all of those details before using the tool.
That is why the organisation needs to establish an approved way to work with AI.
Free tools can come with a different cost
Michael used a familiar principle during his presentation:
If a product is free, the user may be part of the value exchange.
Public AI tools can provide enormous capability without charging the individual user, but businesses still need to understand how the information entered into those tools is handled.
The issue is not that every free AI platform automatically exposes business data.
The issue is that employees may not know which settings apply or whether the account they are using provides appropriate commercial protections.
A personal account and an approved business platform are not necessarily the same thing.
This distinction matters when the user is working with company information.
Businesses should not assume that because a tool is widely used, it is automatically suitable for confidential work.
Why Microsoft Copilot is different inside Microsoft 365
Michael explained that Microsoft Copilot can operate inside the Microsoft 365 environment used by many businesses.
It connects through Microsoft Graph, which allows it to work with approved information across systems such as Outlook, Teams, SharePoint and OneDrive.
That means the AI can help someone find an email, summarise a meeting or work with internal documents without requiring them to manually copy that information into an unrelated public service.
Michael also noted that Microsoft provides a Copilot Chat option across many Microsoft business licences.
This gives employees a protected place to use generative AI for appropriate tasks, with business data handled within the organisation’s Microsoft environment rather than being shared for general model training.
For businesses already using Microsoft 365, this can provide a more controlled starting point.
It does not remove the need for policy, training or good security practices.
It gives the organisation a platform around which those controls can be built.
AI is most valuable when it understands your business
Public generative AI tools are extremely capable when the user wants information drawn from the internet.
They can help with general research, writing and ideas.
The needs of a business are often more specific.
A team may want to build a proposal using previous work completed by the organisation.
They may need to answer an HR question based on the company’s actual policies.
They may want to find a customer conversation from a year ago or create a report using internal data.
For those tasks, the AI needs access to the right business information.
Michael described this as grounding the AI in the organisation’s data.
Rather than generating a generic answer, the tool can use the documents, emails and information the business has approved.
This makes the response more relevant.
It also creates a strong reason to get information governance right.
An AI tool can only work safely with business data when access to that data has already been managed properly.
AI can expose problems that already exist
Many organisations have information stored across Microsoft 365 without fully understanding who can access it.
Files may have been shared broadly in the past.
Old employees may still appear in groups.
Documents may sit in locations that were convenient at the time but are no longer appropriate.
Traditionally, these issues can remain hidden because employees only see the information they know to search for.
AI changes that.
A tool capable of searching across emails, files and collaboration platforms may make it easier to surface information.
That is useful when permissions are correct.
It can be risky when they are not.
If a staff member has access to a folder they should not see, AI may make that information easier to discover.
The AI has not necessarily created the access problem.
It has exposed a weakness that was already present.
This is why businesses should review data access before connecting AI deeply into their systems.
Who can access sensitive documents?
Where are confidential files stored?
Are permissions based on current roles?
What information should an AI agent be able to retrieve?
These questions are part of AI readiness.
Security is the foundation of useful AI
Businesses sometimes view cybersecurity as something that slows innovation down.
In the case of AI, good security is what makes innovation possible.
Clear permissions allow the AI to retrieve useful information without exposing it to the wrong people.
Well-organised data improves the quality of the answers.
Secure identity controls reduce the chance that a compromised account will gain access to AI-powered capabilities.
Approved systems give employees somewhere safe to work instead of pushing them towards unmonitored tools.
The stronger the foundation, the more confidently the organisation can explore agents and automation.
Without that foundation, AI can magnify existing disorder.
Poor data creates poor answers.
Weak access controls create privacy risks.
Unclear policies create inconsistent behaviour.
The tool is only part of the system.
Employees need practical guidance
An AI policy should do more than tell people what they cannot do.
Employees need to know which tools are approved, what kinds of information can be used and when human review is required.
They should understand that customer details, internal financial information and confidential documents cannot simply be pasted into any public platform.
They also need examples.
Can AI be used to rewrite a general marketing paragraph?
Can it summarise a customer contract?
Can it draft an email containing personal information?
Can it produce advice that will be sent without review?
The answers will depend on the organisation, the industry and the tool being used.
Making those expectations clear reduces uncertainty.
Training should also explain that AI can make mistakes.
A confident answer is not necessarily a correct answer.
Employees remain responsible for checking the result before it is used in a business context.
Agents require thoughtful boundaries
AI agents can do more than produce text.
They can retrieve information, create documents and initiate parts of a workflow.
Michael demonstrated agents that collected contact information, built a structured security workshop and identified emails requiring follow-up.
Those capabilities create real value.
They also require sensible boundaries.
What information can the agent access?
What actions can it complete automatically?
Which steps require approval?
How will the results be checked?
A low-risk task such as drafting a meeting summary may be suitable for a high level of automation.
A financial transaction should not be completed based solely on an AI recommendation.
The level of human oversight should reflect the consequence of getting the task wrong.
Businesses need to decide those limits deliberately rather than discovering them after a problem occurs.
Start with a controlled use case
The safest way to introduce AI is not necessarily to make every tool available to everyone on the first day.
A business can begin with one approved use case.
It might create an internal agent that answers questions from a set of HR policies.
It could help a team locate documents inside SharePoint.
It may summarise meeting transcripts and prepare action items for review.
The business can assess whether the information is accurate, whether the permissions work as expected and whether employees find the tool useful.
Lessons from that initial project can then inform the next one.
This allows AI adoption to grow alongside the organisation’s confidence and capability.
It also gives employees a legitimate alternative to using unapproved tools.
The goal is not to stop innovation
Employees use AI because it helps them.
The technology can remove repetitive work, improve access to information and give people a useful first draft.
Trying to eliminate that demand is unlikely to work.
The better approach is to channel it.
Give staff an approved platform.
Explain what good use looks like.
Protect the information the business is responsible for.
Review the systems and permissions underneath it.
Then look for opportunities where AI can genuinely improve the way work is done.
This approach supports innovation without ignoring risk.
Why this matters to us
At eManaged, we regularly speak with businesses that are excited about AI but unsure where to begin.
Some are concerned about security.
Others are worried their employees may already be using tools without approval.
Both concerns are reasonable.
Michael Brook’s Geelong Cyber Sip presentation helped bring those issues together.
AI agents can create meaningful productivity gains, but only when the business understands the information they will use and the controls needed to protect it.
We were delighted to have Michael share his knowledge and practical examples with the Geelong business community.
This is exactly why we hold Cyber Sip events.
We want local businesses to hear about emerging technologies before they are forced to make rushed decisions. We want them to understand both the opportunity and the responsibility that comes with adoption.
Your employees are likely to use AI.
The important question is whether they will use it inside a clear, secure and well-managed environment.
At eManaged, we help organisations assess their technology foundations, strengthen their data security and plan for responsible AI adoption.
Because AI should help your business work better without placing the information you rely on at unnecessary risk.
Talk to eManaged about creating a secure and practical approach to AI in your organisation.