"Safe" depends on the specific tool, the specific plan, and what you paste. Consumer plans and business plans of the same product often have completely different data terms.
The four questions to ask per tool
- Does this plan train models on my inputs? Some consumer tiers default to yes (with opt-outs); business tiers usually default to no. Cursor, for example, states it won't train on your content unless you explicitly agree.
- How long are inputs retained? Retention ranges from zero-retention enterprise commitments to 30-day abuse-review windows to indefinite.
- Do I own the outputs? Most major vendors assign output rights to users, but ownership doesn't guarantee copyright protection for AI-generated content.
- What's prohibited? Client confidential material under NDA may be contractually off-limits regardless of the tool's policy.
The practical rule
Treat every consumer-tier AI chat as a public surface until you've read that plan's training policy. For client work, prefer business/enterprise tiers with explicit no-training commitments, and never paste anything you wouldn't put in an email to a stranger.
Every review on this site links directly to each vendor's privacy documentation so you can verify current terms.