What better way to mark Cybersecurity Awareness Month than by talking about one of today’s most overlooked workplace risks: AI.
AI is now woven into most workplace IT environments. Employees use ChatGPT, Copilot, Gemini, and dozens of other AI tools every day, often without IT’s knowledge. Many use personal accounts and routinely paste sensitive information into systems that sit outside company control.
For MSPs, the challenge is not convincing clients that AI adoption is happening. It’s helping them understand that ungoverned AI adoption creates real security risks. AI may improve productivity, but it can also introduce significant cyber, compliance, and legal exposure.
The rise of shadow AI
The scale of the problem is striking. Microsoft and LinkedIn’s 2024 Work Trend Index found that 78 percent of AI users bring their own AI tools to work. That figure rises to 80 percent among small and midsized businesses.
IBM’s 2025 Cost of a Data Breach Report found that one in five organizations experienced a breach linked to shadow AI. Breaches involving unsanctioned AI tools cost an average of $670,000 more than those that did not. Yet only 37 percent of organizations had policies to manage or detect AI use.
David Matalon, CEO and founder of Venn, sees these risks firsthand. His company provides secure workspace controls for BYOD environments.
Within customer environments monitored by Venn, 62 percent of connections to browser-based AI tools came through personal accounts. After organizations restricted access to company-managed accounts, that number fell to just 5 percent.
“The real risk,” Matalon said, “is which account people sign in with.”
When employees use personal accounts, organizations lose visibility, contractual protections, retention controls, and recourse if something goes wrong.
Small actions can create big risks
The consequences are rarely dramatic at first. Andrew Curtis, founder of CiBRAI and Gadget Access and a veteran vCISO with more than 20 years of experience, frequently encounters the same exposure scenarios.
“Consider a salesperson uploading a customer spreadsheet to prepare a proposal, a finance employee submitting a workbook containing bank details, or a developer pasting a configuration file that includes an API key,” he said.
AI meeting assistants connected to email and calendar systems create additional risk because they continuously access and process organizational data.
“A prompt,” Curtis said, “can be a data transfer disguised as a question.”
These actions often feel harmless to employees. In reality, they can expose sensitive business information, customer records, financial data, credentials, and intellectual property.
The legal consequences many SMBs overlook
Many SMBs have not fully considered the legal implications of AI use.
According to Jasmine Mines, attorney and founder of Mines Law Firm, that knowledge gap creates an opportunity for MSPs to lead the conversation.
“Employees do not think of a prompt as a disclosure, but legally it often is one,” she said. “When someone pastes a customer list, a contract, or a privileged email into a personal AI account, the company has handed that information to an outside party on terms it never negotiated and cannot control.”
Courts have started to address these issues directly. In United States v. Heppner, decided in February 2026, a federal court ruled that communications with a public AI platform are not protected by attorney-client privilege or work-product doctrine. Privileged information entered into a public AI tool can lose those protections.
Trade secrets face similar risks.
“Federal law protects confidential business information only if the owner has taken reasonable measures to keep it secret,” Mines said. “A company that knows employees are feeding source code or pricing models into unvetted tools, and does nothing, may be handing an opposing party the argument that it never took reasonable measures at all.”
For healthcare organizations, entering patient data into a consumer AI tool without a business associate agreement may constitute a HIPAA violation.
Governance works better than bans
All three experts agree that banning AI is an understandable but ineffective response. “A ban does not remove the usage,” Curtis said. “It moves it to a personal account on a personal device, where the organization has no logs, no contract, and no recourse.”
Matalon summarizes the issue clearly: “Why an AI policy that only says no creates more shadow AI.”
Mines makes the business case. “78 percent of AI users bring their own tools to work, and a ban does not end that behavior. It hides it from the people responsible for managing the risk.”
Instead, MSPs should focus on visibility and governance. Curtis recommends first identifying which AI tools employees actually use. That process should combine employee conversations with web, endpoint, application, and identity telemetry. It should also include browser extensions, transcription tools, and AI assistants that many discovery scans overlook. “Ask employees which tools they use and what problems they solve,” he said. “Combine that conversation with available telemetry.”
The resulting inventory is usually much larger than clients expect.
From there, organizations should provide approved AI platforms through company-managed accounts before imposing restrictions. In practice, convenience often influences behavior more than policy.
Curtis recommends selecting a small number of approved services and evaluating their data handling, retention, model-training practices, and integrations before allowing sensitive information to pass through them.
Clear guidelines also matter. Mines recommends a simple rule: credentials, customer records, health information, privileged communications, and source code should never enter an unapproved AI tool. “Give people an approved enterprise tool that is at least as easy as the free one,” she said. “If the secure option is harder, employees will route around it.”
Organizations should also make incident reporting simple. Employees need clear procedures for reporting accidental uploads. Response plans should include credential revocation, containment of connected systems, and investigation of exposed data.
A growing opportunity for MSPs
For MSPs, the business case is clear.
“The goal is not to stop employees from using AI,” Mines said. “It is to make the approved path easier than the unapproved one, so that sensitive information gets the same protection inside an AI tool that it gets everywhere else in the business.”
Curtis frames the MSP role in similarly practical terms.
“MSPs should make safe AI the easiest way to get the job done.”
For clients still evaluating their options, Barracuda AI Data Security provides a practical starting point. The platform helps MSPs identify which AI tools employees use, inspect prompts and file uploads in real time, and produce the audit-ready evidence that insurers and compliance frameworks increasingly require.
The goal is not to block AI. The goal is to govern it before an unlogged, uncontrolled prompt becomes the breach nobody saw coming.
Photo: patpitchaya / Shutterstock
This post originally appeared on Smarter MSP.