What employees should never paste into a public AI chatbot
A practical guide for Hong Kong workers on six categories of data never to paste into public AI chatbots, with redacted examples, legal context, and a pre-paste checklist.
Public AI chatbots such as ChatGPT, Claude, Gemini, and DeepSeek are convenient tools for drafting emails, summarising documents, and generating ideas. But every piece of text you paste into a public chatbot is sent to the provider's servers, often stored, and may be used for model training or reviewed by human moderators. For Hong Kong workers, the risks are not just hypothetical: the city's Personal Data Protection Ordinance (PDPO) and sector-specific regulations can hold both employees and their employers liable for data breaches. Below are six categories of information that should never be pasted into a public AI chatbot, with concrete redacted examples.
1. Personal identifiable information (PII) of colleagues, clients, or the public
Pasting a list of employee names, Hong Kong Identity Card numbers, phone numbers, or email addresses into a chatbot exposes that data to potential leakage. Even if the chatbot provider claims not to store prompts, there is no guarantee that the data will not be accessed by unauthorised parties or used to improve the model.
Redacted example of what not to paste:
- "Please draft a welcome email for our new hire [NAME REDACTED], whose HKID is [HKID REDACTED] and mobile number is [MOBILE REDACTED]."
Safer alternative: Replace all PII with placeholders before pasting. For example: "Draft a welcome email for a new hire named [Employee Name], whose HKID is [HKID] and mobile is [Mobile]."
2. Unreleased financial data, budgets, or sales figures
Quarterly revenue, upcoming product pricing, or confidential budget forecasts are valuable to competitors. If a chatbot provider suffers a breach or if the data is inadvertently exposed through model outputs, your company could suffer financial or reputational damage.
Redacted example of what not to paste:
- "Our Q3 2026 revenue was [AMOUNT REDACTED] HKD, which is [PERCENTAGE REDACTED] above forecast. The board will review the Q4 budget of [AMOUNT REDACTED] next week."
Safer alternative: Use generic terms: "Our Q3 revenue was above forecast. The board will review the Q4 budget next week."
3. Passwords, API keys, or authentication tokens
It may seem obvious, but employees under time pressure have been known to paste a password or API key into a chatbot to ask for help debugging a script. Once pasted, that credential is no longer under your control. It could be stored in logs, used to train future models, or exposed in a data breach.
Redacted example of what not to paste:
- "I can't log into the CRM. My password is [PASSWORD REDACTED]. Can you help?"
Safer alternative: Never paste credentials. Use password managers or reset the password through official channels.
4. Legally privileged communications
Hong Kong recognises legal professional privilege for communications between a lawyer and client. Pasting a draft legal opinion, a settlement offer, or internal communications with the legal team into a public chatbot may waive that privilege, making the documents discoverable in litigation.
Redacted example of what not to paste:
- "Our lawyer [NAME REDACTED] advised that the contract clause on data protection is unenforceable. Please rewrite it to limit our liability."
Safer alternative: Do not paste any text that includes legal advice or attorney-client communications. If you need help drafting, describe the situation in general terms without quoting the legal advice.
5. Customer records or client-specific data
Customer names, account numbers, transaction histories, and support tickets are the lifeblood of many Hong Kong businesses. Pasting them into a chatbot violates the PDPO's data minimisation principle and could lead to regulatory fines or lawsuits.
Redacted example of what not to paste:
- "Customer [NAME REDACTED] with account [ACCOUNT REDACTED] complained about a late delivery on [DATE REDACTED]. Please draft a response."
Safer alternative: Anonymise the record: "A customer complained about a late delivery. Please draft a response."
6. Confidential source material or trade secrets
Internal strategy documents, product roadmaps, proprietary algorithms, and source code are the intellectual property of your employer. Pasting them into a chatbot exposes them to the provider and potentially to the public.
Redacted example of what not to paste:
- "Our new product [PRODUCT NAME REDACTED] uses a proprietary algorithm [ALGORITHM DESCRIPTION REDACTED]. Please write a press release."
Safer alternative: Describe the product in high-level, non-proprietary terms: "Write a press release for a new software product that uses machine learning to improve logistics."
What to do instead: a pre-paste checklist
Before you paste any text into a public AI chatbot, ask yourself these five questions:
- Does this text contain any personal data? If yes, anonymise it or use a placeholder.
- Is this information confidential or proprietary? If yes, do not paste it.
- Could this text harm my employer if it were published? If yes, do not paste it.
- Does my company have an AI usage policy? If yes, follow it. If not, ask your manager or compliance team for guidance.
- Is there a safer way to get the same help? Consider using an internal AI tool, a local model, or simply drafting the text yourself.
Public AI chatbots are powerful, but they are not private. Treat every prompt as if it could be read by your boss, your client, or a regulator. When in doubt, leave it out.
Sources and further reading
- Hong Kong PCPD: Employee generative-AI guidance collection
- Hong Kong PCPD: Model Personal Data Protection Framework
- NIST: Generative AI Profile
Guidance and regulations can change; check the current primary sources before applying them to a consequential workplace decision.