Most employees want to use AI well, but without specific instructions, they don’t know where the line is. The five types of information employees should never enter into AI tools are credentials, personal data, confidential business records, proprietary code, and legal or HR files.
This is especially important now, as AI use has already outpaced guidance. Around 70% of firms now actively use AI, according to a 2026 National Bureau of Economic Research working paper.1 Yet the National Cybersecurity Alliance’s 2026 Oh Behave! Report found that 56% of AI users have received no security or privacy training on AI.2
The same report also found that 32% of workplace AI users have shared sensitive company information with an AI tool without informing their employer. Forty-three percent said their employer hasn’t clearly explained what’s OK to share. Without clear rules, employees make their own. That’s when shadow AI, meaning AI use without company approval, takes hold.
Most of these employees aren’t trying to cut corners. They’re simply trying to work faster, and they need a standard.
Sensitive data in AI prompts is any detail that would harm your company, customers, or employees if it left your control. Unlike the generic warning “Don’t share sensitive data,” this definition gives employees a question to ask themselves before they press enter: Would this text cause harm outside the company? If the answer is yes, keep it to yourself.
Paying for an AI account tier doesn’t change the question. A subscription tier tells an employee which features they can use, not which data the company approved for that tool. Creating an AI acceptable use policy for your company settles that question by explicitly naming approved tools and the data employees may enter into each.
To protect company data from AI misuse, employees should never share certain information with AI tools unless the company approves the tool for that data type.
These categories encompass what not to share with AI tools:
Include what information employees should never enter into AI tools in your company’s training program.
Passwords and API keys give direct access to company systems, so employees should never paste or otherwise enter them into an AI tool. Of all the AI data security risks, exposed credentials are the most direct and the most preventable.
Developers face the clearest version of this problem. When a developer pastes a config file or error log into an AI chat for debugging help, they may inadvertently paste a live key along with it. This is simple to fix, though. Replace real values with placeholders like YOUR_API_KEY, store confidential information in a secrets manager, and if any key has left company control, rotate it.
Customers and employees trusted the company with their personal data for a particular purpose. Pasting that data into an unapproved AI tool means using it for another purpose, and regulations such as GDPR hold your company responsible for how it handles personal data.
Even if the disclosure seems harmless, like summarizing a customer complaint thread, drafting a performance review from notes, or cleaning up a support ticket export, it isn’t. It’s also unnecessary. Employees can get the help they need by removing identifiers first. Replace “Maria Lopez, account 48211” with “Customer A,” and the AI tool performs the same job without information it could leak.
If they realize they or a co-worker have already disclosed sensitive data, employees should report the exposure to the security team immediately.
The security team can then judge whether the exposure triggers customer notice or other obligations.
Treat any employee’s first report as a learning experience, not a disciplinary one. Employees who expect punishment tend to stay quiet. You can’t limit damage you don’t know about.
Good AI training for employees combines a policy that names approved tools with hands-on, role-specific practice using each team’s real work. Tool access doesn’t equate to capability. MIT’s NANDA initiative reported in 2025 that only about 5% of integrated AI pilots delivered measurable returns.3
The difference isn’t the AI. It’s the operator. Michael Burch, Vice President of AI Enablement and Acceleration at Security Journey, says skill and security go together: “You are not capable unless you also understand the security implications of what you are building.”
Security Journey’s AI Advantage platform offers role-based tracks for every level, from foundational AI habits to advanced workflows, department-specific applications, and secure development for non-programmers, with monthly updates and live support from a dedicated customer success manager and support team,.
Start small. Publish the list of information employees should never enter into AI tools, name the approved tools, and give employees a safe way to report mistakes.