By J. Mesa
Your employees use AI chatbots at work. Some use tools you bought. Others use free personal accounts you have never seen. They paste in emails to polish, contracts to summarize, and spreadsheets to analyze. Most mean well. Few of them stop to ask where that text goes. A short written policy fixes the problem without banning a useful tool.
Is it safe to use AI chatbots at work?
Yes, with rules. The tools themselves are legitimate products from large companies. The risk comes from what your staff type into them, which account they use, and how much they trust the output.
What are the risks of using AI at work?
- Data exposure. Text pasted into a chatbot leaves your control. On many free and personal plans, the provider may keep the conversation and use it to train future models.
- Confidentiality and compliance. Patient details, card numbers, and client financial records carry legal duties. A free chatbot account comes with no contract that covers them.
- Wrong answers. AI tools produce confident text that contains errors, invented citations, and made-up numbers. Lawyers have been sanctioned for filing briefs with cases an AI tool invented.
- Account compromise. A chat history holds everything an employee ever pasted. A stolen password hands that history to a criminal.
- Prompt injection. A document, email, or web page can hide instructions that an AI assistant follows. An assistant connected to your inbox could be told to forward messages to a stranger.
- Fake tools. Criminals publish look-alike AI apps and browser extensions that steal data.
- Shadow AI. Staff adopt tools on their own, and you lose track of where company data lives.
Does the AI provider train on my data?
It depends on the product and the plan. As a general pattern:
- Free and personal plans often allow training on your conversations by default. Most offer a setting to turn that off.
- Business, team, and enterprise plans from the major providers state that they do not train on your business data by default, and they offer administrator controls, single sign-on, and data processing agreements.
- API access follows separate terms, usually with no training by default.
Terms change. Read the current privacy policy and data terms for the exact product and plan before you approve it, and record the date you checked.
What should employees never put into an AI chatbot?
Unless the company approved the tool for that kind of data, keep these out:
- Passwords, API keys, and security codes
- Social Security numbers, driver’s license numbers, and dates of birth
- Payment card and bank account numbers
- Patient health information
- Client tax returns and financial records
- Employee records and performance matters
- Contracts and documents under a confidentiality agreement
- Legal advice and privileged communications
- Unreleased financial results and pricing
- Proprietary source code and trade secrets
- Network diagrams, firewall rules, and security reports
A simple test for staff: if you would not post it on a public website, do not paste it into an unapproved AI tool.
What is shadow AI?
Shadow AI is the use of AI tools at work without the company’s knowledge or approval. It includes personal chatbot accounts, AI browser extensions, meeting note-takers that join calls, and AI features that appear inside software you already use.
Banning AI does not end shadow AI. It pushes the use onto personal phones and personal accounts, where you see nothing. Offering an approved tool and clear rules works better.
How do I write an AI acceptable use policy?
Keep it to one or two pages. Cover these points.
- Approved tools. List the AI tools and plans staff may use for work. Everything else needs approval first.
- Accounts. Staff use company accounts for work. No personal accounts for company data.
- Data rules. Sort your data into three groups: public, internal, and restricted. State which groups may go into which tools. Restricted data goes only into tools approved for it in writing.
- Human review. A person checks all AI output for accuracy before it reaches a customer, a regulator, or a decision. The employee who sends it owns it.
- Prohibited uses. No AI for final decisions on hiring, firing, lending, or medical care without human judgment. No impersonation. No use that breaks a law or a client contract.
- Disclosure. State when staff must tell a client or a supervisor that AI helped produce the work.
- New tool requests. Name the person who reviews requests and how fast.
- Meeting recorders. Require consent from all participants and approval for the tool.
- Security. Require multi-factor authentication on AI accounts.
- Reporting. Tell staff to report a mistake, such as pasting restricted data, right away and without fear of punishment.
- Review date. Revisit the policy every six months. This field changes fast.
How do I choose a safe AI tool for my business?
Run it through your vendor review.
- Does the business plan exclude your data from training?
- How long does the provider keep conversations, and can you delete them?
- Does it offer administrator controls, single sign-on, and audit logs?
- Does it hold a SOC 2 Type II report or ISO 27001 certification?
- Will it sign the agreement your industry requires, such as a business associate agreement for HIPAA?
- Where does it store data?
Paying for a business plan for ten employees costs far less than one disclosure of client records through a free account.
What about AI built into the software I already use?
Microsoft 365 Copilot, Google Gemini in Workspace, and AI features in accounting, CRM, and records software work inside your existing accounts. That brings a specific risk: the assistant can reach whatever the user can reach.
Many small businesses share folders more widely than they realize. An assistant makes that visible. An employee who asks about salaries may get an answer drawn from an HR folder that was open to everyone all along. Before you turn these features on, clean up file permissions and confirm that sensitive folders are limited to the people who need them.
What are AI agents, and why do they raise the stakes?
An AI agent does more than answer questions. It takes actions: it reads your email, books meetings, browses websites, fills in forms, and runs tasks across your accounts. An agent that reads untrusted content can be tricked by hidden instructions in that content.
Set limits before you deploy one.
- Give the agent the lowest level of access the task requires.
- Require a human to approve payments, deletions, outgoing messages to customers, and changes to settings.
- Keep agents away from banking and administrator accounts.
- Review the logs of what the agent did.
Can I use AI with patient, card, or financial data?
Only with a tool approved for it and a contract that covers it.
- HIPAA. You need a signed business associate agreement with the AI provider before any patient information goes in. Free consumer tools offer none.
- PCI DSS. Keep card numbers out of AI tools.
- FTC Safeguards Rule. An AI provider that receives customer financial information is a service provider. Assess it, and put security terms in the contract.
How do I train employees on AI?
Add a short module to your security awareness training. Show real examples: a prompt that is fine, a prompt that leaks client data, and an AI answer that contains an invented fact. Teach staff to remove names and identifying details before they paste. Explain how to verify output. Repeat the training when the tools change.
What should I do if someone pasted sensitive data into a chatbot?
- Thank the employee for reporting it.
- Delete the conversation and check the account’s data settings.
- Change any password or key that was exposed.
- Record what data, which tool, which account, and when.
- Ask your attorney whether the event triggers a notification duty.
- Use the event to improve the policy and the training.
Your next step
Ask your staff one question at the next meeting: which AI tools do you use for work? Listen without judgment and write down the answers. That list is the starting point for your policy. Cerberus Cybersecurity writes AI acceptable use policies and trains teams to follow them. See our services or contact us.