For a small business with little or no IT support, blocking AI chatbots on company computers can seem like a straightforward way to protect sensitive information. But an employee who still needs help with a task can pick up a phone, sign in to a personal account, and carry on outside the company’s view.
The business may know the block worked and still have no idea what happened next. If the employee asks a chatbot to draft a reply to a customer complaint, the customer’s name, order details, and private correspondence could all end up in a personal account the company can’t review.
Halis Osman Erkan, CEO of browser security company DefensX, sees a more workable approach for small and midsize businesses. He recommends giving employees access to approved AI tools and explaining what company information they can share when using them.
“It has always been difficult to get employees to comply with cybersecurity policies. This time, it is almost impossible,” he told the E-Commerce Times.
One consequence is what security professionals call shadow AI: using AI tools for work without the company’s knowledge or approval. That can start with an employee entering information into a personal chatbot and extend to giving an outside service ongoing access to a work account.
“Shadow AI usage through unmanaged devices poses one of the greatest risks. An employee connecting their corporate inbox to an uncontrolled AI service creates not only a data exposure risk but also a cybersecurity breach risk that could easily be exploited through AI prompt injection,” he said.
Why AI Is Harder to Block
Blocking unapproved software was easier when employers had more say over which products reached their workers, according to Erkan. A vendor that needed a company’s approval had a reason to meet its security requirements before employees began using the product.
AI developers can reach employees without going through that approval process. A worker can open a chatbot in a browser or download an app to a phone, and some services accept prompts without requiring an account. Even software the company already uses can give employees access to new AI features.
“AI use in the workplace requires guardrails at the endpoint, not road closures that simply won’t work in a 20-lane, interconnected highway system,” Erkan said.
For a small business, that can mean setting rules for what employees share through a company-managed browser. Staff might use an approved chatbot to draft a reply to a customer complaint, for example, provided they remove the customer’s name and order details before submitting the prompt.
When Companies Lose Track of Shared Data
Consider a developer who wants a chatbot’s help troubleshooting a piece of code. If the developer copies proprietary source code into a personal mobile app, the employer may have no record of which service received it or how much code the employee shared.
A similar problem arises if someone pastes customer records into an unapproved tool. Even if the employee later reports what happened, a manager may have little to go on beyond the employee’s recollection. Determining which customers’ information was involved, who had access to the account, and whether colleagues still use the service can become a matter of asking around.
Erkan sees value even in basic monitoring, provided employees understand what the company can see. Knowing which AI services people use gives an employer a starting point for deciding where it needs tighter controls and where staff can continue working under the existing policy.
How DefensX Applies Browser Controls
DefensX’s Secure Enterprise Browser gives businesses a way to set rules for employee use of AI services. According to the company, administrators can limit access to approved tools and require employees to sign in with a work account. An AI provider’s retention policy tells the business how long the service may keep information an employee submits. The employer still has to decide which information employees may submit in the first place.
The browser is only one place where those decisions arise. An employee might go directly to a chatbot in one instance and use an AI assistant inside an office application in another. Both can involve company information, even though only one looks to the employee like a visit to an AI service.
In October, DefensX announced AI Guard One, which the company says extends its browser controls to office extensions, native applications, and autonomous agents. For example, a company that restricts what employees can share with a chatbot may want the same rule to apply when an AI assistant accesses that information through an office application.
“DefensX has deep expertise in understanding and enforcing contextual controls within web browsers. We are extending these capabilities across AI applications and agents at the endpoint, with a unified, single pane of glass for visibility, control, and enforcement,” Erkan said.
What an AI Use Audit Should Cover
An audit of workplace AI use can give a small business a starting point for its acceptable use policy. Three questions can help the business decide what that policy should cover:
- Which AI services are employees using, what accounts do they sign in with, and what kinds of company information do they share?
- Which tools should the business approve, and what information should employees be allowed to enter into each one?
- Where are employees likely to need guidance, and what should they do when a rule prevents them from completing a legitimate task?
According to Erkan, DefensX can produce an initial map of workplace AI use within hours of deployment, showing which accounts employees use, which services they access, and which applications they use to reach them. If personal accounts show up repeatedly, a manager can ask employees what they use those accounts for and whether a company-approved tool would meet the same need.
Match AI Controls to the Task
A recruiter might use the same chatbot to polish a public job posting or to review applicants’ résumés. The posting is already public, while the résumés contain personal information that applicants supplied to the employer. In Erkan’s view, the sensitivity of the material and the purpose of the task should guide the company’s decision about what to allow.
“If AI usage remains within company policies, it should not automatically be considered a risk. Freedom of use comes from understanding which interactions actually pose a threat.”
How the company explains a restriction matters to the employee who encounters it. If a recruiter tries to upload a résumé and sees only that the upload was blocked, the obvious question is how to finish the assignment. A notice explaining that résumés contain personal information and directing the recruiter to an approved hiring tool gives the employee a practical next step.
Employees may have questions about monitoring even when their use of AI falls within company rules. Explaining what the company logs, who can review those records, and where employees can raise concerns gives them a clearer picture of what happens when they use an approved tool.
“Transparency and timely communication with end users are vital to successful governance without creating unnecessary friction,” Erkan said.
