Privacy and sensitive information
The question is not whether these tools are safe. It is what you are permitted to put into them, and what happens to it afterwards.
What you will be able to do
Decide correctly what may go into an assistant, and know where to check when unsure.
Before you start
- None. This one applies to everybody.
Not one policy but several
Data handling varies by provider, by plan and over time. Consumer and business tiers usually differ on retention and on whether content may be used for training. The only reliable answer is the current documentation for the exact account you are using.
Assume nothing carries over from what you read last year.
The categories that need a specific basis
Personal data about identifiable people. Special-category data — health, ethnicity, religion, sexuality, biometrics. Confidential client or employer material. Credentials of any kind. Anything under legal privilege or an NDA.
In the UK these attract specific obligations, and the ICO has published guidance on AI and data protection that is the right starting point for an organisation.
Reduce before you send
Most tasks do not need the identifying details. Replace names with roles, remove account numbers, cut the parts that are not relevant to the question. It is faster than seeking approval and usually loses nothing.
Step by step
- Find the data-handling page for your exact plan and read it.
- Write down what your organisation permits, if anything.
- Before pasting, ask: would I email this to an external supplier?
- Anonymise as a habit, not as an exception.
- Check what any persistent memory feature has retained, and clear it periodically.
What good looks like
- You can state your plan's retention position from having read it.
- Anonymising is automatic rather than remembered.
- Nothing sensitive sits in a persistent memory layer.
What it cannot do
- Provider policies change; check the date on what you read.
- Enterprise agreements may differ substantially from published consumer terms.
When it goes wrong
- Pasting a document with client names to summarise it.
- Replace names with roles first. The summary is unaffected.
- Assuming a workplace tool is approved for all data.
- Approval is usually specific. Check the actual scope.
Privacy and data handling
This guide is general information, not legal advice. Data-protection duties depend on your role, your jurisdiction and what you are processing.
Sources4 referenced
- Guidance on AI and data protection
Information Commissioner's Office · regulatory guidance · link checked 2026-09-06
- Claude developer documentation
Anthropic · official product documentation · link checked 2026-09-06
- OpenAI platform documentation
OpenAI · official product documentation · link checked 2026-09-06
- Guidelines for secure AI system development
National Cyber Security Centre · security guidance · link checked 2026-09-06
Links are checked at build time. If one has moved, tell us.
Read next
Working with files, project instructions and memory
Three ways to give a model persistent context, each with a different lifespan and a different failure mode.
Checked 2026-09-06
Organise files and repeatable workflows safely
File operations are the one place where a mistake is not recoverable by pressing undo. The safety routine is short and worth making automatic.
Checked 2026-09-06
Copyright, provenance, work and opportunity
The legal position is genuinely unsettled in places. The practical position is clearer than the discourse suggests.
Checked 2026-09-06