Using AI safely in your business starts with access, not technology. Who can use it? Which data can the agent see? Which actions require approval? This article offers a practical checklist for businesses that want to use AI without exposing their data.
Before deploying an AI agent, work through these seven points:
Selected access: only employees you designate get access. No public endpoint.
Minimum permissions: each integration gets only the permissions it strictly needs.
Private network access: the agent runs inside your network, not on the open internet.
Model selection: decide which model to use and which data may be sent to it.
Human oversight: high-risk actions require approval before execution.
Backup and recovery: encrypted daily, tested periodically, and stored separately from the node in your network.
Logging: the agent's actions are recorded, so you can review what happened afterward.
At Voltti, these points are the foundation of every setup. Read more on our page about security and privacy.
Even without your own AI agent, you can use AI more safely:
| Measure | What it does | Why it matters |
|---|---|---|
Keep business data out of public chatbots | Do not put confidential data into public tools such as ChatGPT | Otherwise, your data may be stored or used for training |
Your own model provider accounts | Use your own API accounts, not free public versions | You retain control over what happens to your data |
Filter data sent to external models | Decide which data may go to an external model | Prevent sensitive information from leaving your network |
Backups in your network | Back up locally stored data to a separate location in your network, never on the node | A backup lets you restore saved data after a failure or mistake |
Want a secure AI environment? Read more about local AI for businesses or discuss your pilot.
For recovery planning, read what happens to your AI environment if hardware fails.
That depends on how you set it up. Selected access, minimum permissions, and control over which data goes to a model provider make safer use possible. Without these measures, you take unnecessary risks.
No. We do not monitor how you use Hermes and do not have permanent access to your daily work. Our role is installation and support.
The environment is backed up daily. If an incident occurs, we help with recovery and analysis. Specific procedures can be set out in a data processing agreement.
We apply practical data security measures. We discuss specific compliance questions for each situation and can document arrangements in a data processing agreement.
