AI Guides › Trend Watch
By Nigel Guy · 3 min read
When an AI company reports a security incident, most people do one of two things: panic and delete everything, or shrug because "everything leaks eventually." Neither changes much. The panic fades within a week and old habits return; the shrug means the same sensitive material keeps going into the same chat box. A breach is actually a useful prompt — not to react to that one incident, but to fix the habits that decide how much any incident could cost you.
The rule: assume anything you type into an AI tool could one day be exposed, and shape your habits so that if it were, the damage would be small — then a breach becomes an inconvenience rather than a crisis.
Work through these promptly, in this order:
Whether or not you were affected, use the moment to check what you're putting at risk:
| Habit | Question | Better default |
|---|---|---|
| What you paste | Do you paste client data, credentials, or personal details? | Remove or replace identifying details before pasting. |
| Chat history | How long do your conversations stay stored? | Use retention or deletion settings if available; clear old chats you don't need. |
| Training settings | Are your conversations used to train models? | Check and set this deliberately. |
| Connected accounts | What email, drives, or tools have you linked? | Keep only what you use. |
| Account security | Unique password? Two-factor on? | Both, always. |
| Work material | Does your employer allow this tool for this data? | Check the policy before, not after. |
Before sending anything sensitive, ask: "If this conversation appeared in public tomorrow, would it cause real harm to me, a client, or someone else?" If yes, rewrite the input without the sensitive parts, use a tool approved for that data, or don't use AI for that task.