Shadow AI for SMBs : Hidden risks and security tips by Vasily Kolesnikov, Security Expert at Kaspersky

The vast majority of enterprises — including small and medium‑sized businesses — have incorporated AI services into their workflows.

According to an October 2025 survey by the Small Business & Entrepreneurship Council, just 12% report not using AI tools, while most employ a mix of applications across research, marketing, customer service, finance and inventory management.

As a result, we are witnessing the growing challenge of Shadow AI, in which employees use publicly available AI services without approval from their IT department.

This creates uncontrolled data flows and raises the risk of confidential information exposure.

The issue can be especially relevant for medium‑ and small‑sized companies that typically lack a dedicated IT budget for AI service deployment.

So, what are the main threats SMBs may face because of Shadow AI, and how can they mitigate these risks?

Living in the shadows

Some employees find it more convenient to use software that has not been approved by their IT department, and they might well recommend that same software to colleagues.

As a result, half the office staff use unapproved messengers or upload files to possibly unsafe cloud storage services.

This phenomenon is called Shadow IT.

Today, with the growing popularity of AI tools in business workflows, a similar Shadow AI phenomenon is gaining momentum.

Among the main challenges for corporate cybersecurity is the fact that threat actors are increasingly embracing the AI hype and exploit it for their own benefit.

In particular, they actively distribute cyberthreats masquerading as popular AI services.

Meanwhile, employees who are unaware of the potential risks download apps from untrustworthy sites and click on suspicious links.

Between January and April 2026, Kaspersky solutions detected more than 33,300 attacks on small and medium-sized businesses, in which malicious or unwanted software for PCs was disguised as popular AI services.

This figure is almost five times higher than the same period in 2025. At the beginning of 2026 the most common lures in cyberattacks involved malware posing as ChatGPT (42%), Claude (24%) and DeepSeek (20%).

What's more, new AI services appear every day, making it difficult for employees to distinguish a new, unfamiliar service from a fake one.

Cybercriminals also exploit this; for example, recently our experts identified a scam website offering an AI service “built for contractors”.

In reality, victims receive nothing after paying for a subscription, while the scammers keep all the money.

Sharing is not caring

The other issue related to the Shadow AI phenomenon concerns the sharing of confidential corporate information with third party AI chatbots, even legitimate ones.

Thoughtless actions by employees can lead to data breach incidents. Unfortunately, such incidents have already occurred — they may stem from cyberattacks or from vulnerabilities and flaws introduced during AI services’ development process.

For example, some time ago it was discovered that a flaw in ChatGPT allowed some users to view parts of messages from others.

It cannot be ruled out that some of these communications contained corporate data.

It’s also important to remember that any information a user shares with AI chatbots is stored on the developers’ servers and may be used in accordance with that provider’s terms and policies.

Developers of large language model chatbots may use provided data for various purposes, such as training newer versions of their models.

Details about how data is stored and processed are outlined in the Terms of Service, and we strongly advise reviewing these documents before using any AI tool.

Another problem is that credentials (login-password pairs) for various AI services – chatbots, image editing, translation, voice generators, etc. – are being compromised as the services gain popularity.

Cybercriminals typically buy and sell such accounts on dark‑web marketplaces.

These accounts are often initially stolen with data‑stealing malware and then leaked via infostealer log files, where they can be further monetized as valuable assets in the cybercrime ecosystem.

If attackers gain access to an employee's AI service account, they can retrieve the confidential information that the employee uploaded there.

Moreover, they could potentially develop a cyberattack plan against the company using the compromised data.

Leaving the shadows: Cybersecurity tips for SMBs

In general, AI platforms used in a corporate ecosystem should meet high reliability and security requirements.

Protection should be comprehensive: from securing the AI infrastructure to defending against large language model specific attacks (such as indirect prompt injection) and from switching from publicly available cloud services to on‑premise deployments of LLMs.

However, it is understandable that, for small and medium-sized businesses, many of these measures may be impractical, and Shadow AI is difficult to manage because many SMBs often lack sufficient financial and human resources.

Therefore, to minimize cybersecurity risks, SMBs can follow certain cybersecurity best practices: Establish clear usage policies for external services, including AI platforms.

Define procedures for coordinating tasks such as new‑software implementation with the IT department, system administrators and other responsible managers.

Raise security awareness among employees.

Provide dedicated training that teach staff how to detect and address potential cybersecurity threats (malware attacks, phishing, scam, etc.) and track their educational progress.

Avoid downloading applications from unofficial sources; verify that apps are compatible with the operating system that is used on a corporate device.

Teach employees basic safe‑AI practices.

Use strong, unique passwords for AI‑service accounts and change them regularly to reduce the impact of credential theft.

Do not share confidential corporate data with chatbots, as it could become public due to potential data breaches.

In any case, before using any AI service, review its Terms of Service to understand how your data is stored and processed.

Implement cost‑effective cybersecurity solutions that fit your budget, size and industry requirements, emphasizing scalability and easy integration.

Such solutions can detect threats promptly and help mitigate risks, including those stemming from human factors.