In many companies, staff already use ChatGPT, Claude, or Gemini to summarise documents, draft emails, or tidy up reports. The trouble is that nobody really knows what data has been pasted into these tools. A client contract, a payroll sheet, or a customer list can end up with a third-party service after a single copy and paste.
Companies can still use AI safely. Data security when using AI starts with clear rules on which data may be used, who may access it, how activity is recorded, and which settings to check on the vendor's side.
Data security risks when using AI
The first risk is leakage to a third party. Any text entered into an AI service is sent to the provider's servers. Depending on the plan and settings, that text may be stored for a period of time and, on some consumer services, used to train models.
The second risk is access that is too broad. When AI is connected to shared folders or company databases, it can surface information to employees who should not be able to see it. Marketing staff, for instance, should not be able to ask an internal AI assistant about salaries.
The third risk is the lack of a trail. Without a record of who asked what and which data was opened, a company will struggle to trace the source of an incident.
A common example is a clinic admin who pastes a patient list into a free AI tool to put together follow-up schedules faster. The intention is good, yet patient names, phone numbers, and complaints now sit in a service the company never assessed. Incidents like this tend to surface long afterwards, if at all.
Data you should not paste into AI tools
For general AI tools that the company has not configured, use this list as a rule of thumb:
- Personal data of customers or employees, such as national ID numbers (NIK), tax numbers (NPWP), bank account numbers, addresses, and phone numbers.
- Health data, such as clinic patient records.
- Payroll data, performance reviews, and other HR documents.
- Contracts, deeds, and confidential client documents.
- Passwords, API keys, and system login details.
- Unpublished financial reports and internal business plans.
If the work genuinely requires documents like these, use an AI tool the company has approved, with clear data settings and access controls. Another option is to mask sensitive data first, for example replacing NIK numbers and names with codes before the text is sent to a model.
Access control and activity logging
Use company accounts instead of employees' personal accounts. With company accounts, an administrator can decide who may use which AI tools, remove access when someone leaves, and see overall usage. Many business AI services also support sign-in through the company's identity system (SSO).
For AI connected to internal data, permissions should follow your organisation's structure. An AI assistant should only retrieve documents that the person asking is allowed to open. This matters a great deal in notary offices, tax consultancies, and law firms, where each staff member usually handles only some of the clients.
Record the activity in an audit log as well. At a minimum, the log should show the user, the time, which data or tool was accessed, and what was done. Keep the log somewhere ordinary users cannot edit, and assign someone to review it regularly.
Settings on the vendor's side
Before choosing an AI service, read its terms of use and data policy. There are a few things to check.
First, find out whether your data is used to train models and whether that can be switched off. Many services apply different terms to consumer plans than to business plans and API access, so check the terms for the plan you actually use. Second, check how long data is retained, which country the servers are in, and whether the retention period can be shortened.
Third, look at the admin features on offer, such as user management, usage reports, and integration settings. If the AI tool will be connected to Google Drive, OneDrive, or internal systems, make sure you can restrict which folders or data it is allowed to read.
Writing an internal policy
All of the above should be captured in a short AI usage policy that people can easily follow. It should cover the approved AI tools, the types of data that are off limits, the requirement to review AI output before using it, and how to report data that has been pasted into the wrong place.
The policy should be consistent with your company's privacy policy and the principles of Indonesia's Personal Data Protection Law (UU PDP, Law No. 27 of 2022), especially when AI processes personal data of customers or employees. Introduce the policy with real examples from everyday work, and review it whenever a new AI tool comes into use. Name an owner for the policy as well, such as the head of operations or the IT lead, so someone can make the call when new questions come up.
Company AI with data protection built in
AI can work with internal documents and data without opening up more access than necessary. Averoa builds AI over company data with per-team access, answers that cite their sources, and masking of data such as NIK and NPWP before it reaches a model, through our Company AI service.
