Public AI tools can be useful for drafting, summarising and exploring ideas. They should not be treated as a private notebook or a secure place for every kind of business information.
Providers, plans and settings handle data differently, and their terms can change. Before using any AI service, check how prompts and uploaded files are collected, stored, reviewed and used. Share the minimum information required for the task.
1. Passwords, access codes and security secrets
Never paste passwords, passphrases, two-factor authentication codes, recovery codes, API keys, private encryption keys or unrestricted access links into a public AI prompt. A genuine credential should be treated as exposed if it is shared accidentally.
Use a placeholder such as [API KEY] or [ADMIN EMAIL] when you need help understanding a configuration.
2. Customer, staff or supplier personal information
Avoid uploading names, personal email addresses, phone numbers, home addresses, contact lists, identification numbers, signatures, photographs or other information that identifies a person.
The Office of the Australian Information Commissioner recommends, as a matter of best practice, that organisations do not enter personal information—particularly sensitive information—into publicly available generative AI tools because of the privacy risks involved.
3. Health, financial, legal and identity records
Do not enter medical or health information, bank or card details, tax records, payroll information, legal correspondence, passport or driver-licence details, or material connected to a sensitive dispute. Removing a person’s name may not be enough if the remaining details could still identify them.
4. Confidential business and client material
Keep confidential contracts, unpublished pricing, tender information, client briefs, private meeting notes, unreleased campaigns, internal strategy and commercially sensitive financial information out of unapproved public tools.
Check confidentiality agreements and client instructions before using any third-party AI service. Information does not become safe to share merely because the task feels routine.
5. Unpublished intellectual property and licensed assets
Be cautious with source files, original artwork, proprietary methods, unreleased product ideas, source code and licensed photos, fonts or other assets. You may not have permission to upload third-party material, and the provider’s rights and data controls may not suit the work.
Safer ways to ask for help
- Replace real names and details with obvious placeholders.
- Remove information that is not required for the answer.
- Describe the structure of a problem instead of pasting the original confidential document.
- Use a business-approved account, plan and configuration when the work requires stronger data controls.
- Review the provider’s current privacy policy, terms and model-improvement settings.
- Check every output before using or sharing it.
Australian cyber security guidance recommends that small businesses establish rules for what may be shared with AI services, review vendor privacy and data-handling practices, and anonymise personal details where appropriate. Even anonymised information may sometimes be re-identified, so minimisation still matters.
If sensitive information has already been shared
Act promptly. Delete the conversation or uploaded file where the service allows it, but do not assume deletion removes every retained copy. If a password, API key or access code was involved, revoke or rotate it immediately. Record what was shared and who may be affected.
Depending on the information and circumstances, contact the relevant client, your IT or cyber security adviser, and a privacy or legal professional to assess whether any notification or further response is required.
A useful rule: if you would not place the information in an email to an unknown external supplier, do not place it in a public AI prompt.
This article provides general practical information, not legal, privacy or cyber security advice.