AI Safety & Responsible AI
Data leakage
Unintended exposure of private, proprietary, or restricted information.
Example
A private spreadsheet accidentally appears in a response to an unrelated user.
Why it matters
It helps teams recognize when information has reached someone or somewhere it should not.
You'll hear it
“Why did this answer contain another customer's details?”
Can you control it?
Not directly
No, not directly. Data leakage describes a learned property, observed result, or broader concept rather than a normal user setting. You may influence it indirectly by changing the tool, data, prompt, or workflow.
Common questions
- Does leakage always involve a deliberate attack?
- No. Mistakes in permissions, sharing or application behavior can also expose information.
- Can a partial detail still be damaging?
- Yes. A fragment can reveal sensitive facts or help someone infer more.
- Does deleting one visible copy remove every exposure?
- Not necessarily. Other copies, records or recipients may still have the information.