10 AI myths.
What is often misunderstood and what actually applies.
Myth and reality.#
Worth passing on.
| Myth | In fact |
|---|---|
| AI understands what it writes. | Language models calculate likely words. They do not have understanding in the human sense. |
| What AI says has been checked. | AI can invent facts and sources (hallucination). Checking remains a must. |
| Paid versions are automatically privacy-compliant. | What matters are the contract, settings, data location and training on input. |
| My input is private. | That depends on the provider, plan and settings. Many free services analyse input. |
| AI texts automatically belong to me. | Copyright protects human creations. Purely AI-generated content is usually not protected. |
| The AI Act only affects developers. | Companies that use AI also have obligations, for example AI literacy (Art. 4). |
| A ban solves the problem. | Blanket bans often lead to shadow AI. Approved tools and rules work better. |
| You can spot deepfakes immediately. | Fakes are getting better. Rely on a second channel and a code word. |
| Small companies are exempt. | Obligations depend on role and risk. There are simplifications for small companies, but no blanket exemption. |
| A policy takes care of everything. | Training, a register, contracts and regular checks are also part of it. |
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.: "10 AI myths". sicherheitsdenker.de/en/ki-irrtuemer/, as of 6 November 2026.