Understanding AI tools.
Which kind of tool fits which purpose, and what does that mean for your data?
8 types of tool.#
Classified vendor-neutrally.
| Type of tool | Used for | Typical risks | Data |
|---|---|---|---|
| Chatbot and assistant | Texts, ideas, summaries | Data leakage, hallucinations | Yellow, approved tools only |
| Image and video generator | Graphics, drafts, illustrations | Copyright, deepfakes, labelling | No photos of people without a legal basis |
| Transcription and meeting notes | Minutes and summaries | Confidential conversations, consent | Yellow to red, observe consent |
| Translator | Texts and documents | Confidential content at the provider | Yellow, check the contract |
| Coding assistant | Programming and testing | Credentials in code, insecure code | Source code without keys |
| AI search and research | Sources and overview | Wrong or invented sources | Green, check sources |
| Document chat (RAG) | Questions about your own documents | Permissions, poisoned documents | According to the data traffic light per document |
| Agent with tool access | Carries out tasks itself | Excessive permissions, prompt injection | Minimal permissions, approval needed |
Rule of thumb.#
The more permissions and access a tool has, the stricter the rules, approvals and controls must be. Use the data traffic light to classify the data.
Questions? Get in touch.
Write to me if a topic is missing or you want to know more.
Sources and further reading
Was this page helpful?
Found a mistake? Let me know.
Suggested citation: Russo, K.: "Understanding AI tools". sicherheitsdenker.de/en/ki-werkzeuge/, as of 6 November 2026.