Every charger is connected to the internet and accepts commands from the operator's system. This is convenient, but it opens the door to attacks. We explain which threats affect charging stations, what NIS2 changes and how the Polish AI model Bielik helps protect them.
Why charging stations are targets for attacks
A station usually communicates with the operator's system using the OCPP protocol. The same connection carries session and billing data as well as commands: start and stop charging, power changes, restarts and firmware updates. Anyone who takes over this connection, or an account with the right permissions, can stop charging across the whole network or charge at someone else's expense.
The most common threats
- Unencrypted communication. Older OCPP deployments run without encryption, so traffic can be intercepted or modified.
- Charger impersonation. A fake station can send made-up energy readings or obtain free charging.
- Malicious commands. A mass restart, stopped sessions or replaced firmware on many stations at once.
- Compromised accounts. Weak passwords and no second authentication factor in the operator dashboard.
- Billing manipulation. Understated meter readings and sessions that do not match the invoice data.
What NIS2 changes
The NIS2 directive covers, among others, charge point operators in the energy sector. Companies covered by the new rules must manage risk, document events and report serious incidents quickly. The early warning must be sent within 24 hours, the full incident report within 72 hours and the final report within one month. The management board is also accountable for failures. Whether and to what extent NIS2 applies to a given company depends on its size and scale of operations.
A certificate tells you who sent a command. Not whether it makes sense
Encryption and certificates are the foundation, but they are not enough. If an attacker takes over a real account, their commands will look legitimate. So besides checking who sends a command, you need to check whether it makes sense. For example: can a station stop all sessions in the network at once? Is the energy reading physically possible at the power rating of this charge point? Does a restart follow immediately after a configuration change and a firmware update?
This is the job of Powerize Shield. It analyses charger events in real time and holds risky commands from an unknown source until a person approves them. Trusted processes, such as load management and tariff changes, run without interruption.
Where the Bielik AI helps
Bielik is a Polish language model developed by the SpeakLeash foundation and ACK Cyfronet AGH. In Powerize Shield it runs locally, on our own server, not in a foreign cloud. It has three tasks:
- Translates alarms into plain language. Instead of a technical log, you get a short explanation in Polish: what happened, what the impact is and what to do next.
- Helps assess an alarm. It shows whether an event looks like a real attack or a false alarm, and how it may affect ongoing sessions.
- Prepares NIS2 report drafts. The early warning, incident report and final report are ready for review in minutes, instead of being written in a rush.
AI you can trust
A language model can make mistakes, so in Shield it does not make the decisions. Rules and people decide on a block, and dates, station identifiers and figures in reports are inserted by the system, not by the model. Personal data is masked before it reaches the model, and every event is recorded in a tamper-evident log that will stand up to an audit.
Where to start
- Check that your stations connect to the system over an encrypted connection.
- Enable two-factor authentication for all accounts in the operator dashboard.
- Decide who reports an incident, how they do it and where you keep the evidence.
- Check whether and to what extent NIS2 applies to you.