What should you not tell GPT?

What should you not tell ChatGPT: Privacy Risks

Users of what should you not tell ChatGPT must handle data carefully to avoid identity theft and phishing attacks. Understanding how to manage your privacy settings when using AI ensures your personal information remains secure. Learn how to protect your identity by controlling the data you share during your AI interactions.

What should you not tell ChatGPT?

Deciding what information to share with AI requires a cautious approach because input data may be used for model training purposes. As a rule of thumb, you should never enter anything you would not feel comfortable posting publicly. This protects you from potential data leakage and ensures your sensitive information remains secure.

Protecting Personally Identifiable Information

Personally Identifiable Information (PII) is the most critical data to withhold.^[1] Avoid sharing your full name, home address, social security numbers, passport details, or private phone numbers in any prompt. These details are sensitive and, if exposed, can significantly increase your privacy risks of using AI chatbots.

I remember the first time I almost pasted a project document containing my home address; the realization that this data could live in a training set indefinitely was a serious wake-up call. Now, I always sanitize my inputs by replacing specific identifiers with generic placeholders like User1 or Address_Placeholder before starting a session.

Handling Financial and Medical Data

Financial data such as credit card numbers, bank routing codes, and investment credentials should never be shared with AI chatbots. Similarly, sensitive medical history, including specific diagnoses, prescriptions, or laboratory results, must stay off these platforms. Most public AI models are not designed to be HIPAA-compliant, meaning your health information lacks the security protocols required for clinical data.

Securing Business Secrets and Credentials

Confidential business information, such as unreleased financial reports, proprietary strategies, or private client lists, presents a major risk if uploaded to a public AI. Sharing these internal documents could lead to the unintentional leaking of trade secrets. Additionally, never use the platform as a digital vault for what data to keep private from AI.

Understanding Enterprise Privacy Settings

If you are using an Enterprise or Plus account, you often have more control over your privacy settings. You can typically disable model training to prevent your chat history from being used to update the underlying AI model.^[2] Adjusting these preferences is a smart move for anyone using AI for professional workflows and does ChatGPT store your conversation data.

Data Sharing Risks

Publicly Available Data

- General news, historical facts, public speeches

- Low - information already exists in the public domain

Sensitive Personal Data

- Passwords, SSNs, medical records, financial credentials

- High - potential for identity theft or fraud

Minh's Corporate Security Lesson

Minh, a software engineer at a tech firm in Ho Chi Minh City, was trying to debug a complex database query. Under pressure to meet a deadline, he copied a snippet of his company's internal code that contained hardcoded API keys.

He pasted the code directly into a public AI chatbot. Almost immediately, he felt a knot in his stomach—the realization that he had just exposed private credentials hit him hard.

He quickly revoked the keys and forced a rotation through his security team. It was an exhausting 24-hour cycle of panic and cleanup that taught him to never paste code without cleaning it first.

Since that day, Minh strictly adheres to a policy of scrubbing all sensitive data from snippets before using any AI tools. He now uses environment variables instead of hardcoded keys, ensuring his work remains secure.

This information is for educational purposes only and does not replace professional security advice. Always consult your organization's IT security policies regarding the use of AI tools for sensitive work data.