ChatGPT at Work, Artificial Intelligence, and Information Security

AI and Security Insights

ChatGPT at Work and Information Security

For many organisations, artificial intelligence has been treated as a productivity issue, a governance issue, or a compliance issue. Far fewer have treated it as what it now clearly is: an information security issue.

ChatGPT and similar AI tools do not sit neatly inside existing security architectures. They operate outside traditional perimeter controls, process information dynamically, and are often used through interfaces that bypass established monitoring and logging mechanisms. When these tools enter the workplace, they introduce new data flows that many security models were not designed to handle.

This page explains why AI use at work creates information security risk, how that risk actually materialises, and why restricting access alone often makes the problem worse.

Why AI changes the information security landscape

Traditional information security models are built around systems, networks, and endpoints that organisations can see and control. AI tools disrupt this model.

When employees use AI tools, information leaves controlled environments and is processed by external systems in ways that are not always transparent. Prompts may contain fragments of sensitive data. Outputs may be stored, reused, or shared. The organisation’s ability to track, classify, or contain that information is significantly reduced.

This does not mean AI tools are inherently insecure. It means they introduce new attack surfaces and leakage pathways that many organisations have not yet accounted for.

The misconception of control through restriction

A common response to AI-related security concerns is to restrict access. Organisations block public AI tools, limit access to internal solutions, or prohibit use through policy. In practice, this often increases risk.

When people need AI to do their jobs and approved tools are unavailable or unevenly distributed, they route around controls. They use personal devices. They rely on public tools outside corporate environments. They copy, paste, screenshot, or photograph information to work around restrictions.

From an information security perspective, this is the worst outcome. Activity moves from monitored systems into unmonitored ones. Logging disappears. Controls are bypassed. The organisation loses any realistic ability to detect or respond to misuse.

Why uneven access creates the greatest exposure

Some organisations deploy internal or custom AI tools but restrict access based on role, employment status, or licensing. Permanent staff may have access. Contractors, third parties, or offshore teams may not.

This creates two parallel environments. One governed and visible. One invisible and unmanaged.

Security risk concentrates in the unmanaged environment.

The organisation may believe it has mitigated AI risk by building internal tools, while in reality it has displaced the most dangerous behaviour into areas it cannot see. This creates a false sense of security that only becomes apparent after an incident or during external scrutiny.

What effective organisations do differently

Organisations that manage AI-related information security risk start with visibility rather than restriction. They focus on understanding how AI is actually used across the workforce, including contractors and third parties. They align access, guidance, and monitoring with real workflows rather than idealised ones.

They also treat AI use as part of their existing information security framework, rather than a separate novelty. AI risk is assessed alongside other data processing activities, with clear ownership and documentation.

This approach does not eliminate risk, but it prevents it from becoming invisible.

How this fits with governance and compliance

Information security, AI governance, and data protection are inseparable in practice. An organisation cannot credibly claim to govern AI if it does not understand the security implications of how information flows through AI tools.

This is why AI risk assessment, governance, and information security must be addressed together rather than in isolation. Each depends on the same underlying visibility.

When this requires immediate attention

This issue should be addressed urgently if AI use is believed to be restricted but cannot be evidenced, if contractors or third parties are involved in sensitive work, or if security teams cannot confidently describe how AI-related data exposure is monitored.

In these situations, the absence of incidents should not be taken as reassurance. It often indicates a lack of detection rather than a lack of risk.

Next steps

If your organisation is concerned about how AI use intersects with information security, the first step is understanding what is actually happening across systems, teams, and access boundaries.

An AI risk assessment provides the visibility needed to address information security risk in a controlled and defensible way.

AI use as a data leakage problem

From an information security standpoint, the most significant AI risk is not model behaviour. It is data leakage.

Sensitive information can be exposed through prompts, attachments, outputs, or downstream reuse. Because AI interactions are conversational and informal, users may not recognise when they are sharing more than intended. Traditional data loss prevention controls are often poorly suited to this context.

Once information has been processed through external AI systems, containment becomes extremely difficult. This shifts the security conversation from prevention to damage limitation.

Why information security teams struggle with AI

Many security teams find AI uncomfortable because it does not fit existing threat models cleanly. AI use is often fragmented, user-driven, and opaque. It crosses boundaries between IT, data protection, governance, and operations.

Security teams are then asked to “secure AI” without clarity on where it is being used, by whom, or for what purpose. Without visibility, security controls become speculative.

This is not a failure of security teams. It is a structural challenge created by the way AI enters organisations.