Your employees are using ChatGPT, Copilot, Claude, or whatever launched last week. Some are pasting customer data, code, financial figures, or patient records into these tools. The question isn't whether this is happening. It is. The question is whether you have architecture controls around it.
Each of these sends data to a third-party service. Some of that data is personal data under GDPR. Once it's sent, you can't get it back.
Data exposure: Whatever goes into the prompt may be stored, used for training, or exposed. If an employee pastes customer PII into ChatGPT, that's a data breach.
Compliance violation: If regulated data leaves your controlled environment through an LLM, that's a potential breach of NIS2, DORA, GDPR, or sector-specific rules.
IP leakage: Source code, trade secrets, internal processes. Once they're in an external AI service, you've lost control of them.
Banning AI tools doesn't work. People use them anyway. The control is architecture — sanctioned tools, DLP, logging — not a policy that says "don't."
Aethel Guard is a Netherlands-based security architecture firm serving EU companies of all sizes. Your architect, not your vendor.