8 Dinge, die du ChatGPT, Claude & Co niemals verraten solltest

Summary of Key Advice: 8 Things to Never Share With ChatGPT/Claude/AI Tools

1) Don’t use AI as a diary or therapist

• AI isn’t a human with real emotional intelligence.
• It doesn’t provide confidentiality like a lawyer or doctor would.
• Deep feelings and “dark secrets” can end up stored on servers and potentially be reviewed by humans.

2) Don’t paste sensitive ID / financial / account data

• Never enter ID numbers, bank details, passwords, or similar sensitive information into prompts.
• Data sent to AI is stored externally and may be impossible to delete.
• Take a few extra seconds to remove sensitive lines before submitting.

3) Don’t upload portraits / biometric data

• Avoid uploading your face, your children’s photos, or biometric identifiers (especially high-resolution portraits).
• Facial features are hard to “change” later (unlike passwords).
• Uploading biometrics can enable identity tracking/recognition by third parties.

4) Don’t use AI for medical diagnoses or interpreting lab results

• AI can only infer from patterns—it can’t examine you or provide clinical judgment.
• It may overestimate danger (causing fear) or downplay symptoms (leading to delayed care).
• Your medical info is sensitive and should be handled by real healthcare professionals.

5) Don’t rely on AI for legal matters or contracts

• Legal systems don’t tolerate mistakes.
• AI can hallucinate, fabricating legal rules, deadlines, or case-like references with high confidence.
• Treat AI output as non-authoritative and verify it with a qualified lawyer.

6) Don’t paste secrets into AI during coding

• Never copy code that includes API keys, JWT tokens, database passwords, or secret keys.
• These can end up in provider logs and/or training pipelines.
• Golden rule: replace sensitive strings with placeholders before pasting code.

7) Don’t share proprietary company code or trade secrets

• Avoid pasting internal algorithms or business logic into public AI chats.
• There’s a risk of legal exposure unless an enterprise contract guarantees conditions like zero/instant data retention.
• Some companies block AI tools specifically due to data leakage risks.

8) Don’t reveal system architecture or your system prompts

• Don’t share infrastructure details (ports, IP structure, database bindings) or internal system prompt text.
• This can effectively give attackers a “map” of potential vulnerabilities.
• Keep infrastructure blueprints and system prompts within your own systems.

Wellness / Mindfulness Angle (Implied)

• Pause and protect boundaries: The message emphasizes taking a short moment to avoid impulsive sharing—especially when stressed, anxious, or in a hurry.
• Don’t outsource sensitive decisions: Emotional, medical, legal, and security-critical decisions should not be externalized to systems without true confidentiality and accountability.

Presenters / Sources

• Su (Model My Mind): Video creator and presenter.
• Example referenced source (unnamed): A US lawyer discussing an AI chatbot being used against a defendant in court (exact person not named in the subtitles).