Remote work has established a model that is here to stay. Along with it comes an expanded attack surface that most companies have not yet fully mapped. When an employee accesses corporate systems from home, a coffee shop, or another country, they are doing more than simply using a different network. They are making decisions about company data in an environment where security controls differ, distractions are greater, and the tools they use may not have been approved by the security team. The arrival of generative AI tools has added a new layer to this problem. Employees who use ChatGPT, Copilot, Gemini, or any other assistant to optimize their work often paste excerpts from documents, customer data, or strategic information into these tools without realizing the risks involved. In this article, you will understand why remote access security has become more complex with the use of AI, what the main exposure vectors are, and what your company needs to have in place to control this environment. Enjoy the read.
What Has Changed in Remote Access Security
For a long time, protecting remote access basically meant ensuring that the connection was encrypted. A corporate VPN addressed most of the problem: data traveled securely between the employee’s device and the company’s servers.
This model is still necessary, but it is no longer sufficient.
The problem is that today, data does not leave the company only through the network. It also leaves through employee behavior, the tools they choose to use, and the decisions they make beyond the reach of corporate monitoring. An employee working from home who pastes a contract into an AI tool to “summarize it before the meeting” has just sent that document to an external server, possibly in another country, without anyone at the company being notified.
The Specific Risks of AI Tools in Remote Work
Generative AI tools have become part of many employees’ workflows, and this has happened faster than security policies could keep up. The result is a gap between what employees do in their daily routines and what the company knows is being done with its data.
This gap has real consequences.
Data Leaking Through Prompts
When an employee uses a public AI tool to analyze a document, process a spreadsheet, or draft a communication based on internal data, that information is sent to the tool’s server. Depending on the terms of use and account settings, this content may be used to train future models or retained in logs accessible to the provider.
For customers’ personal data, this represents a violation of Brazil’s General Data Protection Law (LGPD). For strategic business information, it represents an exposure that no agreement with the tool can undo once the data has left the company.
Unapproved Tools on Personal Devices
Employees working from home frequently switch between personal and corporate devices or use browser extensions and applications that have never undergone an evaluation by the IT team. An AI extension installed in a personal browser may have access to everything the employee types, including credentials and data from internal systems accessed through that same browser.
This type of exposure does not appear in any corporate log because it happens outside the company’s perimeter.
Personal Accounts Used with Corporate AI Tools
Even tools approved by the company can become a problem when employees use personal accounts instead of corporate accounts. Personal accounts may have different privacy settings, may not be subject to the company’s data retention policies, and may remain active after an employee leaves the company, with corporate data still stored in them.
The Controls That Need to Be in Place
Remote access security in environments where AI is used cannot be addressed with a single tool. It requires a combination of technical controls, clear policies, and effective training that work together.
An AI Usage Policy for Corporate Data
The starting point is defining what employees can and cannot do with AI tools when their work involves company data. Which tools are allowed, under what conditions, with what types of data, and through which accounts.
Without a documented and clearly communicated policy, each employee makes whatever decision seems reasonable to them. And what seems reasonable to someone trying to be productive can be very different from what is reasonable for someone managing data risks. An information security policy that does not address the use of generative AI is outdated compared to the actual working environment of most teams.
Context-Based Access Control
The traditional access control model checks whether an employee has permission to access a particular system. The more modern model also considers the context: which device is being used, which network the connection comes from, what time it occurs, and whether the behavior is typical.
An access attempt from an unknown device, in an unusual location, or outside regular business hours should trigger an alert, regardless of whether the credentials are correct. This is the principle of zero trust applied to remote access: no access is trusted by default, regardless of its origin.
Monitoring Data Movement
Knowing that employees have access to systems is not enough. Companies need to know what is being done with data during those access sessions. An employee downloading a volume of files far above their historical pattern within 20 minutes is a signal that needs to be investigated, whether because they are about to leave the company or because their account has been compromised.
Data Loss Prevention (DLP) solutions that monitor file transfers outside the corporate environment are particularly relevant in remote work models, where employees have greater autonomy and less direct supervision than they would in an office.
Multifactor Authentication for All Critical Access
Compromised credentials are the most common entry point in incidents involving remote access. A strong password is not enough when an employee reuses passwords across multiple services, connects through inadequately secured networks, or becomes the target of a well-crafted social engineering attack.
Multifactor authentication does not eliminate the risk, but it significantly increases the effort required to exploit compromised credentials. For systems containing sensitive data, it is the minimum protection that must be enabled.
What Does the LGPD Require in This Scenario?
Remote work and the use of AI do not create an exception to LGPD obligations. The company remains responsible for the personal data it processes, regardless of where the employee accessing that data is working or which tools they are using to process it.
If an employee working from home uses a public AI tool to process customers’ personal data without this activity being included in the company’s data mapping and without an adequate legal basis, the company — not the employee — is responsible for the resulting exposure.
This means that data mapping must include remote data flows and the AI tools employees use. It also means that the incident notification protocol to the Brazilian National Data Protection Authority (ANPD) must account for scenarios in which a data breach occurs outside the company’s controlled technical environment.
How to Start Structuring These Controls
The first step is understanding what is actually happening. Which AI tools employees use for work, which accounts they use, and what types of data they share. This information is rarely documented and often comes as a surprise to managers when it is finally uncovered.
Based on this assessment, the company can determine which tools will be approved for use with corporate accounts, which will be blocked, and which will require additional controls. It can also structure technical remote access controls based on what the team’s actual working environment requires, rather than on what was anticipated when working from home was still an exception.
STWBrasil helps companies establish security controls for remote work environments, including AI usage policies, access monitoring, and data leak prevention. If your company has not yet mapped how remote employees are using corporate data, that is the place to start.
